Math Archives

Tag Archive for: Math

Automatic Financial Trading Agent for Low-risk Portfolio Management using Deep Reinforcement Learning

June 9, 2022/in Artificial Intelligence, Data Science, Deep Learning, Machine Learning, Main Category/by Jieyun Hu

This article focuses on autonomous trading agent to solve the capital market portfolio management problem. Researchers aim to achieve higher portfolio return while preferring lower-risk actions. It uses deep reinforcement learning Deep Q-Network (DQN) to train the agent. The main contribution of their work is the proposed target policy.

Introduction

Author emphasizes the importance of low-risk actions for two reasons: 1) the weak positive correlation between risk and profit suggests high returns can be obtained with low-risk actions, and 2) customer satisfaction decreases with increases in investment risk, which is undesirable. Author challenges the limitation of Supervised Learning algorithm since it requires domain knowledge. Thus, they propose Reinforcement Learning to be more suitable, because it only requires state, action and reward specifications.

The study verifies the method through the back-test in the cryptocurrency market because it is extremely volatile and offers enormous and diverse data. Agents then learn with shorter periods and are tested for the same period to verify the robustness of the method.

2 Proposed Method

The overall structure of the proposed method is shown below.

The architecutre of the proposed trading agent system.

2.1 Problem Definition

The portfolio consists of m assets and one base currency.

The price vector p stores the price p of all assets:

The portfolio vector w stores the amount of each asset:

At time $𝑡$ , the total value $W_t$ of the portfolio is defined as the inner product of the price vector $p_t$ and the portfolio vector $w_t$ .

Finally, the goal is to maximize the profit $P_t$ at the terminal time step $𝑇$ .

2.2 Asset Data Preprocessing

1) Asset Selection
Data is drawn from the Binance Exchange API, where top m traded coins are selected as assets.

2) Data Collection
Each coin has 9 properties, shown in Table.1, so each trade history matrix has size (α * 9), where α is the size of the target period converted into minutes.

3) Zero-Padding
Pad all other coins to match the matrix size of the longest coin. (Coins have different listing days)

Comment: Author pointed out that zero-padding may be lacking, but empirical results still confirm their method covering the missing data well.

4) Stack Matrices
Stack m matrices of size (α * 9) to form a block of size (m* α * 9). Then, use sliding window method with widow size w to create (α – w + 1) number of sequential blocks with size (w * m * 9).

5) Normalization
Normalize blocks with min-max normalization method. They are called history block 𝜙 and used as input (ie. state) for the agent.

3. Deep Q-Network

The proposed RL-based trading system follows the DQN structure.

Deep Q-Network has 2 networks, Q- and Target network, and a component called experience replay. The Q-network is the agent that is trained to produce the optimal state-action value (aka. q-value).

Comment: Q-value is calculated by the Bellman equation, which, in short, consists of the immediate reward from next action, and the discounted value of the next state by following the policy for all subsequent steps.

Here,
Agent: Portfolio manager
Action a: Trading strategy according to the current state
State 𝜙 : State of the capital market environment
Environment: Has all trade histories for assets, return reward r and provide next state 𝜙’ to agent again

DQN workflow:

DQN gets trained in multiple time steps of multiple episodes. Let’s look at the workflow of one episode.

Training of a Deep Q-Network

1) Experience replay selects an action according to the behavior policy, executes in the environment, returns the reward and next state. This experience set ( $\phi_t, a_t, r_r,\phi_{t+!}$ ) is stored in the repository as a sample of training data.

2) From the repository of prior observations, take a random batch of samples as the input to both Q- and Target network. The Q-network takes the current state and action from each data sample and predicts the q-value for that particular action. This is the ‘Predicted Q-Value’.Comment: Author uses 𝜀-greedy algorithm to calculate q-value and select action. To simplify, 𝜀-greedy policy takes the optimal action if a randomly generated number is greater than 𝜀, which represents a tradeoff between exploration and exploitation.

The Target network takes the next state from each data sample and predicts the best q-value out of all actions that can be taken from that state. This is the ‘Target Q-Value’.

Comment: Author proposes a different target policy to calculate the target q-value.

3) The Predicted q-value, Target q-value, and the observed reward from the data sample is used to compute the Loss to train the Q-network.

Comment: Target Network is not trained. It is held constant to serve as a stable target for learning and will be updated with a frequency different from the Q-network.

4) Copy Q-network weights to Target network after n time steps and continue to next time step until this episode is finished.

4.0 Main Contribution of the Research

4.1 Action and Reward

Agent determines not only action a but ratio , at which the action is applied.

Action:
Hold, buy and sell. Buy and sell are defined discretely for each asset. Hold holds all assets. Therefore, there are (2m + 1) actions in the action set A.

Agent obtains q-value of each action through q-network and selects action by using 𝜀-greedy algorithm as behavior policy.
Ratio:
$\sigma$ is defined as the softmax value for the q-value of each action (ie. i-th asset at $\sigma = 0.5$ , then i-th asset is bought using 50% of base currency).
Reward:
Reward depends on the portfolio value before and after the trading strategy. It is clipped to [-1,1] to avoid overfitting.

4.2 Proposed Target Policy

Author sets the target based on the expected SARSA algorithm with some modification.

Comment: Author claims that greedy policy ignores the risks that may arise from exploring other outcomes other than the optimal one, which is fatal for domains where safe actions are preferred (ie. capital market).

The proposed policy uses softmax algorithm adjusted with greediness according to the temperature term 𝜏. However, softmax value is very sensitive to the differences in optimal q-value of states. To stabilize learning, and thus to get similar greediness in all states, author redefine 𝜏 as the mean of absolute values for all q-values in each state multiplied by a hyperparameter 𝜏’.

4.3 Q-Network Structure

This study uses Convolutional Neural Network (CNN) to construct the networks. Detailed structure of the networks is shown in Table 2.

Comment: CNN is a deep neural network method that hierarchically extracts local features through a weighted filter. More details see: https://towardsdatascience.com/stock-market-action-prediction-with-convnet-8689238feae3.

5 Experiment and Hyperparameter Tuning

5.1 Experiment Setting

Data is collected from August 2017 to March 2018 when the price fluctuates extensively.

Three evaluation metrics are used to compare the performance of the trading agent.

Profit $P_t$ introduced in 2.1.
Sharpe Ratio: A measure of return, taking risk into account.

Comment: $p_t$ is the standard deviation of the expected return and $P_f$ is the return of a risk-free asset, which is set to 0 here.
Maximum Drawdown: Maximum loss from a peak to a through, taking downside risk into account.

5.2 Hyperparameter Optimization

The proposed method has a number of hyperparameters: window size mentioned in 2.2, 𝜏’ in the target policy, and hyperparameters used in DQN structure. Author believes the former two are key determinants for the study and performs GridSearch to set w = 30, 𝜏’ = 0.25. The other hyperparameters are determined using heuristic search. Specifications of all hyperparameters are summarized in the last page.

Comment: Heuristic is a type of search that looks for a good solution, not necessarily a perfect one, out of the available options.

5.3 Performance Evaluation

Benchmark algorithms:

UBAH (Uniform buy and hold): Invest in all assets and hold until the end.
UCRP (Uniform Constant Rebalanced Portfolio): Rebalance portfolio uniformly for every trading period.

Methods from other studies: hyperparameters as suggested in the studies
EG (Exponential Gradient)
PAMR (Passive Aggressive Mean Reversion Strategy)

Comment: DQN basic uses greedy policy as the target policy.

The proposed DQN method exhibits the best overall results out of the 6 methods. When the agent is trained with shorter periods, although MDD increases significantly, it still performs better than benchmarks and proves its robustness.

6 Conclusion

The proposed method performs well compared to other methods, but there is a main drawback. The encoding method lacked a theoretical basis to successfully encode the information in the capital market, and this opaqueness is a rooted problem for deep learning. Second, the study focuses on its target policy, while there remains room for improvement with its neural network structure.

Specification of Hyperparameters.

References

Shin, S. Bu and S. Cho, “Automatic Financial Trading Agent for Low-risk Portfolio Management using Deep Reinforcement Learning”, https://arxiv.org/pdf/1909.03278.pdf
Li, P. Zhao, S. C. Hoi, and V. Gopalkrishnan, “PAMR: passive aggressive mean reversion strategy for portfolio selection,” Machine learning, vol. 87, pp. 221-258, 2012.
P. Helmbold, R. E. Schapire, Y. Singer, and M. K. Warmuth, “On‐line portfolio selection using multiplicative updates,” Mathematical Finance, vol. 8, pp. 325-347, 1998.

https://deepai.org/machine-learning-glossary-and-terms/softmax-layer#:~:text=The%20softmax%20function%20is%20a,can%20be%20interpreted%20as%20probabilities.

http://www.kasimte.com/2020/02/14/how-does-temperature-affect-softmax-in-machine-learning.html

https://towardsdatascience.com/reinforcement-learning-made-simple-part-2-solution-approaches-7e37cbf2334e

https://towardsdatascience.com/reinforcement-learning-explained-visually-part-4-q-learning-step-by-step-b65efb731d3e

https://towardsdatascience.com/reinforcement-learning-explained-visually-part-3-model-free-solutions-step-by-step-c4bbb2b72dcf

https://towardsdatascience.com/reinforcement-learning-explained-visually-part-5-deep-q-networks-step-by-step-5a5317197f4b

On the difficulty of language: prerequisites for NLP with deep learning

January 27, 2021/in Artificial Intelligence, Big Data, Data Mining, Data Science, Data Science Hack, Deep Learning, Machine Learning, Main Category, Natural Language Processing, Python, TensorFlow, Text Mining/by Yasuto Tamura

1 Preface

This section is virtually just my essay on language. You can skip this if you want to get down on more technical topic.

As I do not study in natural language processing (NLP) field, I would not be able to provide that deep insight into this fast changing deep leaning field throughout my article series. However at least I do understand language is a difficult and profound field, not only in engineering but also in many other study fields. Some people might be feeling that technologies are eliminating languages, or one’s motivations to understand other cultures. First of all, I would like you to keep it in mind that I am not a geek who is trying to turn this multilingual world into a homogeneous one and rebuild Tower of Babel, with deep learning. I would say I am more keen on social or anthropological sides of language.

I think you would think more about languages if you have mastered at least one foreign language. As my mother tongue is Japanese, which is totally different from many other Western languages in terms of characters and ambiguity, I understand translating is not what learning a language is all about. Each language has unique characteristics, and I believe they more or less influence one’s personalities. For example, many Western languages make the verb, I mean the conclusion, of sentences clear in the beginning part of the sentences. That is also true of Chinese, I heard. However in Japanese, the conclusion comes at the end, so that is likely to give an impression that Japanese people are being obscure or indecisive. Also, Japanese sentences usually omit their subjects. In German as well, the conclusion of a sentences tend to come at the end, but I am almost 100% sure that no Japanese people would feel German people make things unclear. I think that comes from the structures of German language, which tends to make the number, verb, relations of words crystal clear.

Let’s take an example to see how obscure Japanese is. A Japanese sentence 「頭が赤い魚を食べる猫」can be interpreted in five ways, depending on where you put emphases on.

Common sense tells you that the sentence is likely to mean the first two cases, but I am sure they can mean those five possibilities. There might be similarly obscure sentences in other languages, but I bet few languages can be as obscure as Japanese. Also as you can see from the last two sentences, you can omit subjects in Japanese. This rule is nothing exceptional. Japanese people usually don’t use subjects in normal conversations. And when you read classical Japanese, which Japanese high school students have to do just like Western students learn some of classical Latin, the writings omit subjects much more frequently.

*However interestingly we have rich vocabulary of subjects. The subject “I” can be translated to 「私」、「僕」、「俺」、「自分」、「うち」etc, depending on your personality, who you are talking to, and the time when it is written in.

I believe one can see the world only in the framework of their language, and it seems one’s personality changes depending on the language they use. I am not sure whether the language originally determines how they think, or how they think forms the language. But at least I would like you to keep it in mind that if you translate a conversation, for example a random conversation at a bar in Berlin, into Japanese, that would linguistically sound Japanese, but not anthropologically. Imagine that such kind of random conversation in Berlin or something is like playing a catch, I mean throwing a ball named “your opinion.” On the other hand, normal conversations of Japanese people are in stead more of, I would say, “resonance” of several tuning forks. They do their bests to show that they are listening to each other, by excessively nodding or just repeating “Really?”, but usually it seems hardly any constructive dialogues have been made.

*I sometimes feel you do not even need deep learning to simulate most of such Japanese conversations. Several-line Python codes would be enough.

My point is, this article series is mainly going to cover only a few techniques of NLP in deep learning field: sequence to sequence model (seq2seq model) , and especially Transformer. They are, at least for now, just mathematical models and mappings of a small part of this profound field of language (as far as I can cover in this article series). But still, examples of language would definitely help you understand Transformer model in the long run.

2 Tokens and word embedding

*Throughout my article series, “words” just means the normal words you use in daily life. “Tokens” means more general unit of NLP tasks. For example the word “Transformer” might be denoted as a single token “Transformer,” or maybe as a combination of two tokens “Trans” and “former.”

One challenging part of handling language data is its encodings. If you started learning programming in a language other than English, you would have encountered some troubles of using keyboards with different arrangements or with characters. Some comments on your codes in your native languages are sometimes not readable on some software. You can easily get away with that by using only English, but when it comes to NLP you have to deal with this difficulty seriously. How to encode characters in each language should be a first obstacle of NLP. In this article we are going to rely on a library named BPEmb, which provides word embedding in various languages, and you do not have to care so much about encodings in languages all over the world with this library.

In the first section, you might have noticed that Japanese sentence is not separated with spaces like Western languages. This is also true of Chinese language, and that means we need additional tasks of separating those sentences at least into proper chunks of words. This is not only a matter of engineering, but also of some linguistic fields. Also I think many people are not so conscious of how sentences in their native languages are grammatically separated.

The next point is, unlike other scientific data, such as temperature, velocity, voltage, or air pressure, language itself is not measured as numerical data. Thus in order to process language, including English, you first have to map language to certain numerical data, and after some processes you need to conversely map the output numerical data into language data. This section is going to be mainly about one-hot encoding and word embedding, the ways to convert word/token into numerical data. You might already have heard about this

You might have learnt about word embedding to some extent, but I hope you could get richer insight into this topic through this article.

2.1 One-hot encoding

One-hot encoding would be the most straightforward way to encode words/tokens. Assume that you have a dictionary whose size is $|\mathcal{V}|$ , and it includes words from “a”, “ablation”, “actually” to “zombie”, “?”, “!”

In a mathematical manner, in order to choose a word out of those $|\mathcal{V}|$ words, all you need is a $|\mathcal{V}|$ dimensional vector, one of whose elements is $1$ , and the others are $0$ . When you want to choose the No. i word, which is “indeed” in the example below, its corresponding one-hot vector is $\boldsymbol{v} = (0, \dots, 1, \dots, 0 )$ , where only the No. i element is $1$ . One-hot encoding is also easy to understand, and that’s all. It is easy to imagine that people have already come up with more complicated and better way to encoder words. And one major way to do that is word embedding.

2.2 Word embedding

Source: Francois Chollet, Deep Learning with Python,(2018), Manning

Actually word embedding is related to one-hot encoding, and if you understand how to train a simple neural network, for example densely connected layers, you would understand word embedding easily. The key idea of word embedding is denoting each token with a $D$ dimensional vector, whose dimension is fewer than the vocabulary size $|\mathcal{V}|$ . The elements of the resulting word embedding vector are real values, I mean not only 0 or 1. Obviously you can encode much richer variety of tokens with such vectors. The figure at the left side is from “Deep Learning with Python” by François Chollet, and I think this is an almost perfect and simple explanation of the comparison of one-hot encoding and word embedding. But the problem is how to get such convenient vectors. The answer is very simple: you have only to train a network whose inputs are one-hot vector of the vocabulary.

The figure below is a simplified model of word embedding of a certain word. When the word is input into a neural network, only the corresponding element of the one-hot vector is 1, and that virtually means the very first input layer is composed of one neuron whose value is 1. And the only one neuron propagates to the next D dimensional embedding layer. These weights are the very values which most other study materials call “an embedding vector.”

When you input each word into a certain network, for example RNN or Transformer, you map the input one-hot vector into the embedding layer/vector. The examples in the figure are how inputs are made when the input sentences are “You’ve got the touch” and “You’ve got the power.” Assume that you have a dictionary of one-hot encoding, whose vocabulary is {“the”, “You’ve”, “Walberg”, “touch”, “power”, “Nights”, “got”, “Mark”, “Boogie”}, and the dimension of word embeding is 6. In this case $|\mathcal{V}| = 9, D=6$ . When the inputs are “You’ve got the touch” or “You’ve got the power” , you put the one-hot vector corresponding to “You’ve”, “got”, “the”, “touch” or “You’ve”, “got”, “the”, “power” sequentially every time step $t$ .

In order to get word embedding of certain vocabulary, you just need to train the network. We know that the words “actually” and “indeed” are used in similar ways in writings. Thus when we propagate those words into the embedding layer, we can expect that those embedding layers are similar. This is how we can mathematically get effective word embedding of certain vocabulary.

More interestingly, if word embedding is properly trained, you can mathematically “calculate” words. For example, $\boldsymbol{v}_{king} - \boldsymbol{v}_{man} + \boldsymbol{v}_{woman} \approx \boldsymbol{v}_{queen}$ , $\boldsymbol{v}_{Japan} - \boldsymbol{v}_{Tokyo} + \boldsymbol{v}_{Vietnam} \approx \boldsymbol{v}_{Hanoi}$ .

*I have tried to demonstrate this type of calculation on several word embedding, but none of them seem to work well. At least you should keep it in mind that word embedding learns complicated linear relations between words.

I should explain word embedding techniques such as word2vec in detail, but the main focus of this article is not NLP, so the points I have mentioned are enough to understand Transformer model with NLP examples in the upcoming articles.

3 Language model

Language models is one of the most straightforward, but crucial ideas in NLP. This is also a big topic, so this article is going to cover only basic points. Language model is a mathematical model of the probabilities of which words to come next, given a context. For example if you have a sentence “In the lecture, he opened a _.”, a language model predicts what comes at the part “_.” It is obvious that this is contextual. If you are talking about general university students, “_” would be “textbook,” but if you are talking about Japanese universities, especially in liberal art department, “_” would be more likely to be “smartphone. I think most of you use this language model everyday. When you type in something on your computer or smartphone, you would constantly see text predictions, or they might even correct your spelling or grammatical errors. This is language modelling. You can make language models in several ways, such as n-gram and neural language models, but in this article I can explain only general formulations for such models.

*I am not sure which algorithm is used in which services. That must be too fast changing and competitive for me to catch up.

As I mentioned in the first article series on RNN, a sentence is usually processed as sequence data in NLP. One single sentence is denoted as $\boldsymbol{X} = (\boldsymbol{x}^{(1)}, \dots, \boldsymbol{x}^{(\tau)})$ , a list of vectors. The vectors are usually embedding vectors, and the $(t)$ is the index of the order of tokens. For example the sentence “You’ve go the power.” can be expressed as $\boldsymbol{X} = (\boldsymbol{x}^{(1)}, \boldsymbol{x}^{(2)}, \boldsymbol{x}^{(3)}, \boldsymbol{x}^{(4)})$ , where $\boldsymbol{x}^{(1)}, \boldsymbol{x}^{(2)}, \boldsymbol{x}^{(3)}, \boldsymbol{x}^{(4)}$ denote “You’ve”, “got”, “the”, “power”, “.” respectively. In this case $\tau = 4$ .

In practice a sentence $\boldsymbol{X}$ usually includes two tokens $BOS$ and $EOS$ at the beginning and the end of the sentence. They mean “Beginning Of Sentence” and “End Of Sentence” respectively. Thus in many cases $\boldsymbol{X} = (\boldsymbol{BOS} , \boldsymbol{x}^{(1)}, \dots, \boldsymbol{x}^{(\tau)}, \boldsymbol{EOS} )$ . $\boldsymbol{BOS}$ and $\boldsymbol{EOS}$ are also both vectors, at least in the Tensorflow tutorial.

$P(\boldsymbol{X} =$ $(\boldsymbol{BOS}, \boldsymbol{x}^{(1)}, \dots, \boldsymbol{x}^{(\tau)}, \boldsymbol{EOS})$ is the probability of incidence of the sentence. But it is easy to imagine that it would be very hard to directly calculate how likely the sentence $\boldsymbol{X}$ appears out of all possible sentences. I would rather say it is impossible. Thus instead in NLP we calculate the probability $P(\boldsymbol{X})$ as a product of the probability of incidence or a certain word, given all the words so far. When you’ve got the words $(\boldsymbol{x}^{(1)}, \dots, \boldsymbol{x}^{(t-1})$ so far, the probability of the incidence of $\boldsymbol{x}^{(t)}$ , given the context is $P(\boldsymbol{x}^{(t)}|\boldsymbol{x}^{(1)}, \dots, \boldsymbol{x}^{(t-1)})$ . $P(\boldsymbol{BOS})$ is a probability of the the sentence $\boldsymbol{X}$ being $(\boldsymbol{BOS})$ , and the probability of $\boldsymbol{X}$ being $(\boldsymbol{BOS}, \boldsymbol{x}^{(1)})$ can be decomposed this way: $P(\boldsymbol{BOS}, \boldsymbol{x}^{(1)}) =$ $P(\boldsymbol{x}^{(1)}|\boldsymbol{BOS})P(\boldsymbol{BOS})$ .

Just as well $P(\boldsymbol{BOS}, \boldsymbol{x}^{(1)}, \boldsymbol{x}^{(2)}) =$ $P(\boldsymbol{x}^{(2)}| \boldsymbol{BOS}, \boldsymbol{x}^{(1)}) P( \boldsymbol{BOS}, \boldsymbol{x}^{(1)})=$ $P(\boldsymbol{x}^{(2)}| \boldsymbol{BOS}, \boldsymbol{x}^{(1)}) P(\boldsymbol{x}^{(1)}| \boldsymbol{BOS}) P( \boldsymbol{BOS})$ .

Hence, the general probability of incidence of a sentence $\boldsymbol{X}$ is $P(\boldsymbol{X})=P(\boldsymbol{BOS}, \boldsymbol{x}^{(1)}, \boldsymbol{x}^{(2)}, \dots, \boldsymbol{x}^{(\tau -1)}, \boldsymbol{x}^{(\tau)}, \boldsymbol{EOS}) =$ $P(\boldsymbol{EOS}| \boldsymbol{BOS}, \boldsymbol{x}^{(1)}, \dots, \boldsymbol{x}^{(\tau)})$ $P(\boldsymbol{x}^{(\tau)}| \boldsymbol{BOS}, \boldsymbol{x}^{(1)}, \dots, \boldsymbol{x}^{(\tau - 1)})$ $\cdots$ $P(\boldsymbol{x}^{(2)}| \boldsymbol{BOS}, \boldsymbol{x}^{(1)})$ $P(\boldsymbol{x}^{(1)}| \boldsymbol{BOS})$ $P(\boldsymbol{BOS})$ .

Let $\boldsymbol{x}^{(0)}$ be $\boldsymbol{BOS}$ and $\boldsymbol{x}^{(\tau + 1)}$ be $\boldsymbol{EOS}$ . Plus, let $P(\boldsymbol{x}^{(t+1)}|\boldsymbol{X}_{[0, t]})$ be $P(\boldsymbol{x}^{(t+1)}|\boldsymbol{x}^{(0)}, \dots, \boldsymbol{x}^{(t)})$ , then $P(\boldsymbol{X}) = P(\boldsymbol{x}^{(0)})\prod_{t=0}^{\tau}{P(\boldsymbol{x}^{(t+1)}|\boldsymbol{X}_{[0, t]})}$ . Language models calculate which words to come sequentially in this way.

Here’s a question: how would you evaluate a language model?

I would say the answer is, when the language model generates words, the more confident the language model is, the better the language model is. Given a context, when the distribution of the next word is concentrated on a certain word, we can say the language model is confident about which word to come next, given the context.

*For some people, it would be more understandable to call this “entropy.”

Let’s take the vocabulary {“the”, “You’ve”, “Walberg”, “touch”, “power”, “Nights”, “got”, “Mark”, “Boogie”} as an example. Assume that $P(\boldsymbol{X}) =$ $P(\boldsymbol{BOS}, \boldsymbol{You've}, \boldsymbol{got}, \boldsymbol{the}, \boldsymbol{touch}, \boldsymbol{EOS}) =$ $P(\boldsymbol{BOS}, \boldsymbol{x}^{(1)}, \boldsymbol{x}^{(2)}, \boldsymbol{x}^{(3)}, \boldsymbol{x}^{(4)}, \boldsymbol{EOS})=$ $P(\boldsymbol{x}^{(0)})\prod_{t=0}^{4}{P(\boldsymbol{x}^{(t+1)}|\boldsymbol{X}_{[0, t]})}$ . Given a context $(\boldsymbol{BOS}, \boldsymbol{x}^{(1)})$ , the probability of incidence of $\boldsymbol{x}^{(2)}$ is $P(\boldsymbol{x}^{2}|\boldsymbol{BOS}, \boldsymbol{x}^{(1)})$ . In the figure below, the distribution at the left side is less confident because probabilities do not spread widely, on the other hand the one at the right side is more confident that next word is “got” because the distribution concentrates on “got”.

*You have to keep it in mind that the sum of all possible probability $P(\boldsymbol{x}^{(2)} | \boldsymbol{BOS}, \boldsymbol{x}^{(1)})$ is $1$ , that is, $P(\boldsymbol{the}| \boldsymbol{BOS}, \boldsymbol{x}^{(1)}) +$ $P(\boldsymbol{You've}| \boldsymbol{BOS}, \boldsymbol{x}^{(1)}) + \cdots$ $+ P(\boldsymbol{Boogie}| \boldsymbol{BOS}, \boldsymbol{x}^{(1)}) = 1$ .

While the language model generating the sentence “BOS You’ve got the touch EOS”, it is better if the language model keeps being confident. If it is confident, $P(\boldsymbol{X})=$ $P(\boldsymbol{BOS})$ $P(\boldsymbol{x}^{(1)}|\boldsymbol{BOS}}P(\boldsymbol{x}^{(3)}|\boldsymbol{BOS}, \boldsymbol{x}^{(1)}, \boldsymbol{x}^{(2)})$ $P(\boldsymbol{x}^{(4)}|\boldsymbol{BOS}, \boldsymbol{x}^{(1)}, \boldsymbol{x}^{(2)}, \boldsymbol{x}^{(3)})$ $P(\boldsymbol{EOS}|\boldsymbol{BOS}, \boldsymbol{x}^{(1)}, \boldsymbol{x}^{(2)}, \boldsymbol{x}^{(3)}, \boldsymbol{x}^{(4)})}$ gets higher. Thus $(-1) \{ log_{b}{P(\boldsymbol{BOS})} +$ $log_{b}{P(\boldsymbol{x}^{(1)}|\boldsymbol{BOS}}) +$ $log_{b}{P(\boldsymbol{x}^{(3)}|\boldsymbol{BOS}, \boldsymbol{x}^{(1)}, \boldsymbol{x}^{(2)})} +$ $log_{b}{P(\boldsymbol{x}^{(4)}|\boldsymbol{BOS}, \boldsymbol{x}^{(1)}, \boldsymbol{x}^{(2)}, \boldsymbol{x}^{(3)})} +$ $log_{b}{P(\boldsymbol{EOS}|\boldsymbol{BOS}, \boldsymbol{x}^{(1)}, \boldsymbol{x}^{(2)}, \boldsymbol{x}^{(3)}, \boldsymbol{x}^{(4)})} \}$ gets lower, where usually $b=2$ or $b=e$ .

This is how to measure how confident language models are, and the indicator of the confidence is called perplexity. Assume that you have a data set for evaluation $\mathcal{D} = (\boldsymbol{X}_1, \dots, \boldsymbol{X}_n, \dots, \boldsymbol{X}_{|\mathcal{D}|})$ , which is composed of $|\mathcal{D}|$ sentences in total. Each sentence $\boldsymbol{X}_n =$ $(\boldsymbol{x}^{(0)})\prod_{t=0}^{\tau ^{(n)}}{P(\boldsymbol{x}_{n}^{(t+1)}|\boldsymbol{X}_{n, [0, t]})}$ has $\tau^{(n)}$ tokens in total excluding $\boldsymbol{BOS}, \boldsymbol{EOS}$ . And let $|\mathcal{V}|$ be the size of the vocabulary of the language model. Then the perplexity of the language model is $b^z$ , where $z = \frac{-1}{|\mathcal{V}|}\sum_{n=1}^{|\mathcal{D}|}{\sum_{t=0}^{\tau ^{(n)}}{log_{b}P(\boldsymbol{x}_{n}^{(t+1)}|\boldsymbol{X}_{n, [0, t]})}$ . The $b$ is usually $2$ or $e$ .

For example, assume that $\mathcal{V}$ is vocabulary {“the”, “You’ve”, “Walberg”, “touch”, “power”, “Nights”, “got”, “Mark”, “Boogie”}. Also assume that the evaluation data set for perplexity of a language model is $\mathcal{D} = (\boldsymbol{X}_1, \boldsymbol{X}_2)$ , where $\boldsymbol{X_1} =(\boldsymbol{You've}, \boldsymbol{got}, \boldsymbol{the}, \boldsymbol{touch})$ $\boldsymbol{X_2} = (\boldsymbol{You've}, \boldsymbol{got}, \boldsymbol{the }, \boldsymbol{power})$ . In this case $|\mathcal{V}|=9,$ $|\mathcal{D}|=2$ . I have already showed you how to calculate the perplexity of the sentence “You’ve got the touch.” above. You just need to do a similar thing on another sentence “You’ve got the power”, and then you can get the perplexity of the language model.

*If the network is not properly trained, it would also be confident of generating wrong outputs. However, such network still would give high perplexity because it is “confident” at any rate. I’m sorry I don’t know how to tackle the problem. Please let me put this aside, and let’s get down on Transformer model soon.

Appendix

Let’s see how word embedding is implemented with a very simple example in the official Tensorflow tutorial. It is a simple binary classification task on IMDb Dataset. The dataset is composed to comments on movies by movie critics, and you have only to classify if the commentary is positive or negative about the movie. For example when you get you get an input “To be honest, Michael Bay is a terrible as an action film maker. You cannot understand what is going on during combat scenes, and his movies rely too much on advertisements. I got a headache when Mark Walberg used a Chinese cridit card in Texas. However he is very competent when it comes to humorous scenes. He is very talented as a comedy director, and I have to admit I laughed a lot.“, the neural netowork has to judge whether the statement is positive or negative.

This networks just takes an average of input embedding vectors and regress it into a one dimensional value from 0 to 1. The shape of embedding layer is (8185, 16). Weights of neural netowrks are usually implemented as matrices, and you can see that each row of the matrix corresponds to emmbedding vector of each token.

*It is easy to imagine that this technique is problematic. This network virtually taking a mean of input embedding vectors. That could mean if the input sentence includes relatively many tokens with negative meanings, it is inclined to be classified as negative. But for example, if the sentence is “This masterpiece is a dark comedy by Charlie Chaplin which depicted stupidity of the evil tyrant gaining power in the time. It thoroughly mocked Germany in the time as an absurd group of fanatics, but such propaganda could have never been made until ‘Casablanca.'” , this can be classified as negative, because only the part “masterpiece” is positive as a token, and there are much more words with negative meanings themselves.

The official Tensorflow tutorial provides visualization of word embedding with Embedding Projector, but I would like you to take more control over the data by yourself. Please just copy and paste the codes below, installing necessary libraries. You would get a map of vocabulary used in the text classification task. It seems you cannot find clear tendency of the clusters of the tokens. You can try other dimension reduction methods to get maps of the vocabulary by for example using Scikit Learn.

import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import layers
import tensorflow_datasets as tfds
tfds.disable_progress_bar()

(train_data, test_data), info = tfds.load(
    'imdb_reviews/subwords8k', 
    split = (tfds.Split.TRAIN, tfds.Split.TEST), 
    with_info=True, as_supervised=True)

train_batches = train_data.shuffle(1000).padded_batch(10)
test_batches = test_data.shuffle(1000).padded_batch(10)

embedding_dim=16

encoder = info.features['text'].encoder

model = keras.Sequential([
  layers.Embedding(encoder.vocab_size, embedding_dim),
  layers.GlobalAveragePooling1D(),
  layers.Dense(16, activation='relu'),
  layers.Dense(1)
])

print("\n\nThe size of the vocabulary generated from IMDb Dataset is " + str(encoder.vocab_size) + '\n\n')

model.summary()

model.compile(optimizer='adam',
              loss=tf.keras.losses.BinaryCrossentropy(from_logits=True),
              metrics=['accuracy'])

history = model.fit(
    train_batches,
    epochs=10,
    validation_data=test_batches, validation_steps=20)

word_embedding_vectors = model.layers[0].get_weights()[0]

print("\n\nThe shape of the trained weigths of the embedding layer is " + str(word_embedding_vectors.shape) + '\n\n')

from sklearn.manifold import TSNE
X_reduced = TSNE(n_components = 2, init='pca', random_state=0).fit_transform(word_embedding_vectors)

import numpy as np
embedding_dict = zip(encoder.subwords, np.arange(len(encoder.subwords)))
embedding_dict = dict(embedding_dict)

import matplotlib.pyplot as plt

plt.figure(figsize=(60, 45))
plt.scatter(X_reduced[:, 0], X_reduced[:, 1])

for i in range(0, len(encoder.subwords), 5):
    plt.text(X_reduced[i, 0], X_reduced[i, 1], encoder.subwords[i], fontsize=20, color='red')
plt.title("The map of vocabulary of IMDb Dataset mapped to a 2 dimensional space by t-SNE", fontsize=60)
#plt.savefig('imdb_tsne_map.png')
plt.show()

Rethinking linear algebra: visualizing linear transformations and eigen vectors

October 27, 2020/in Data Mining, Data Science, Machine Learning, Main Category, Mathematics/by Yasuto Tamura

In terms of calculation processes of Principal Component Analysis (PCA) or Linear Discriminant Analysis (LDA), which are the dimension reduction techniques I am going to explain in the following articles, diagonalization is what they are all about. Throughout this article, I would like you to have richer insight into diagonalization in order to prepare for understanding those basic dimension reduction techniques.

When our professor started a lecture on the last chapter of our textbook on linear algebra, he said “It is no exaggeration to say that everything we have studied is for this ‘diagonalization.'” Until then we had to write tons of numerical matrices and vectors all over our notebooks, calculating those products, adding their rows or columns to other rows or columns, sometimes transposing the matrices, calculating their determinants.

It was like the scene in “The Karate Kid,” where the protagonist finally understood the profound meaning behind the prolonged and boring “wax on, wax off” training given by Miyagi (or “jacket on, jacket off” training given by Jackie Chan). We had finally understood why we had been doing those seemingly endless calculations.

Source: http://thinkbedoleadership.com/secret-success-wax-wax-off/

But usually you can do those calculations easily with functions in the Numpy library. Unlike Japanese college freshmen, I bet you are too busy to reopen textbooks on linear algebra to refresh your mathematics. Thus I am going to provide less mathematical and more intuitive explanation of diagonalization in this article.

1, The mainstream ways of explaining diagonalization.

*The statements below are very rough for mathematical topics, but I am going to give priority to offering more visual understanding on linear algebra in this article. For further understanding, please refer to textbooks on linear algebra. If you would like to have minimum understandings on linear algebra needed for machine learning, I recommend the Appendix C of Pattern Recognition and Machine Learning by C. M. Bishop.

In most textbooks on linear algebra, the explanations on dioagonalization is like this (if you are not sure what diagonalization is or if you are allergic to mathematics, you do not have to read this seriously):

Let $V (dimV = D)$ be a vector space and let $T_A : V \rightarrow V$ be a mapping of $V$ into itself, defined as $T_A(v) = A \cdot \boldsymbol{v}$ , where $A$ is a $D\times D$ matrix and $\boldsymbol{v}$ is $D$ dimensional vector. An element $\boldsymbol{v} \in V$ is called an eigen vector if there exists a number $\lambda$ such that $A \cdot \boldsymbol{v}= \lambda \cdot \boldsymbol{v}$ and $\boldsymbol{v} \neq \boldsymbol{0}$ . In this case $\lambda$ is uniquely determined and is called an eigen value of $A$ belonging to the eigen vector $\boldsymbol{v}$ .

Any matrix $A$ has $D$ eigen values $\lambda_{i}$ , belonging to $\boldsymbol{v}_{i} (i=1, 2, …., D)$ . If $\boldsymbol{v}_{i}$ is basis of the vector space $V$ , then $A$ is diagonalizable.

When $A$ is diagonalizable, with $D \times D$ matrices $P = (\boldsymbol{v}_{1}, \dots, \boldsymbol{v}_{D})$ , whose column vectors are eigen vectors $\boldsymbol{v}_{i} (i=1, 2, …., D)$ , the following equation holds: $P^{-1}AP = \Lambda$ , where $\Lambda = diag(\lambda_{1}, \dots, \lambda_{D})= \begin{pmatrix} \lambda_{1} & 0& \ldots &0\\ 0 & \lambda_{2} & \ldots & 0 \\ \vdots & \vdots & \ddots & \vdots \\ 0 & 0 & \ldots & \lambda_{D} \end{pmatrix}$ .

And when $A$ is diagonalizable, you can diagonalize $A$ as below.

Most textbooks keep explaining these type of stuff, but I have to say they lack efforts to make it understandable to readers with low mathematical literacy like me. Especially if you have to apply the idea to data science field, I believe you need more visual understanding of diagonalization. Therefore instead of just explaining the definitions and theorems, I would like to take a different approach. But in order to understand them in more intuitive ways, we first have to rethink waht linear transformation $T_A$ means in more visible ways.

2, Linear transformations

Even though I did my best to make this article understandable to as little prerequisite knowledge, you at least have to understand linear transformation of numerical vectors and with matrices. Linear transformation is nothing difficult, and in this article I am going to use only 2 or 3 dimensional numerical vectors or square matrices. You can calculate linear transformation of $\boldsymbol{v}$ by A as equations in the figure. In other words, $\boldsymbol{u}$ is a vector transformed by $A$ .

*I am not going to use the term “linear transformation” in a precise way in the context of linear algebra. In this article or in the context of data science or machine learning, “linear transformation” for the most part means products of matrices or vectors.

*Forward/back propagation of deep learning is mainly composed of this linear transformation. You keep linearly transforming input vectors, frequently transforming them with activation functions, which are for the most part not linear transformation.

As you can see in the equations above, linear transformation with A transforms a vector to another vector. Assume that you have an original vector $\boldsymbol{v}$ in grey and that the vector $\boldsymbol{u}$ in pink is the transformed $\boldsymbol{v}$ by A is. If you subtract $\boldsymbol{v}$ from $\boldsymbol{u}$ , you can get a displacement vector, which I displayed in purple. A displacement vector means the transition from a vector to another vector.

Let’s calculate the displacement vector with more vectors $\boldsymbol{v}$ . Assume that $A =\begin{pmatrix} 3 & 1 \\ 1 & 2 \end{pmatrix}$ , and I prepared several grid vectors $\boldsymbol{v}$ in grey as you can see in the figure below. If you transform those grey grid points with $A$ , they are mapped into the vectors $\boldsymbol{u}$ in pink. With those vectors in grey or pink, you can calculate the their displacement vectors $\boldsymbol{u} = \boldsymbol{v}$ in purple.

I think you noticed that the displacement vectors in the figure above have some tendencies. In order to see that more clearly, let’s calculate displacement vectors with several matrices $A$ and more grid points. Assume that you have three $2 \times 2$ square matrices $A_1 =\begin{pmatrix} 3 & 1 \\ 1 & 2 \end{pmatrix}, A_2 =\begin{pmatrix} 3 & 1 \\ -1 & 1 \end{pmatrix}, A_3 =\begin{pmatrix} 1 & -1 \\ 1 & 1 \end{pmatrix}$ , and I plotted displace vectors made by the matrices respectively in the figure below.

I think you noticed some characteristics of the displacement vectors made by those linear transformations: the vectors are swirling and many of them seem to be oriented in certain directions. To be exact, some displacement vectors have extend in the same directions as some of original vectors in grey. That means linear transformation by $A$ did not change the direction of the original vector $\boldsymbol{v}$ , and the unchanged vectors are called eigen vectors. Real eigen vectors of each A are displayed as arrows in yellow in the figure above. But when it comes to $A_3$ , the matrix does not have any real eigan values.

In linear algebra, depending on the type matrices $A$ , you have consider various cases such as whether the matrices have real or imaginary eigen values, whether the matrices are diagonalizable, whether the eigen vectors are orthogonal, or whether they are unit vectors. But those topics are out of the scope of this article series, so please refer to textbooks on linear algebra if you are interested.

Luckily, however, in terms of PCA or LDA, you only have to consider a type of matrices named positive semidefinite matrices, which $A_1$ is classified to, and I am going to explain positive semidefinite matrices in the fourth section.

3, Eigen vectors as coordinate system

Source: Ian Stewart, “Professor Stewart’s Cabinet of Mathematical Curiosities,” (2008), Basic Books

Let me take Fibonacci numbers as an example to briefly see why diagonalization is useful. Fibonacci is sequence is quite simple and it is often explained using an example of pairs of rabbits increasing generation by generation. Let $a_n (n=0, 1, 2, …)$ be the number of pairs of grown up rabbits in the $n^{th}$ generation. One pair of grown up rabbits produce one pair of young rabbit The concrete values of $a_n$ are $a_0 = 0$ , $a_1 = 1$ , $a_2=1$ , $a_3=2$ , $a_4=3$ , $a_5=5$ , $a_6=8$ , $a_7=13, \dots$ . Assume that $A =\begin{pmatrix} 1 & 1 \\ 1 & 0 \end{pmatrix}$ and that $\begin{pmatrix} a_1 \\ a_0 \end{pmatrix} =\begin{pmatrix} 1 \\ 0 \end{pmatrix}$ , then you can calculate the number of the pairs of grown up rabbits in the next generation with the following recurrence relation. $\begin{pmatrix} a_{n+1} \\ a_{n} \end{pmatrix}=\begin{pmatrix} 1 & 1 \\ 1 & 0 \end{pmatrix} \cdot \begin{pmatrix} a_{n+1} \\ a_{n} \end{pmatrix}$ .Let $\boldsymbol{a}_n$ be $\begin{pmatrix} a_{n+1} \\ a_{n} \end{pmatrix}$ , then the recurrence relation can be written as $\boldsymbol{a}_{n+1} = A \boldsymbol{a}_n$ , and the transition of $\boldsymbol{a}_n$ are like purple arrows in the figure below. It seems that the changes of the purple arrows are irregular if you look at the plots in normal coordinate.

Assume that $\lambda _1, \lambda_2 (\lambda _1< \lambda_2)$ are eigen values of $A$ , and $\boldsymbol{v}_1, \boldsymbol{v}_2$ are eigen vectors belonging to them respectively. Also let $\alpha, \beta$ scalars such that $\begin{pmatrix} a_{1} \\ a_{0} \end{pmatrix} = \begin{pmatrix} 1 \\ 0 \end{pmatrix} = \alpha \boldsymbol{v}_1 + \beta \boldsymbol{v}_2$ . According to the definition of eigen values and eigen vectors belonging to them, the following two equations hold: $A\boldsymbol{v}_1 = \lambda_1 \boldsymbol{v}_1, A\boldsymbol{v}_2 = \lambda_2 \boldsymbol{v}_2$ . If you calculate $\boldsymbol{a}_1$ is, using eigen vectors of $A$ , $\boldsymbol{a}_1 = A\boldsymbol{a}_0 = A (\alpha \boldsymbol{v}_1 + \beta \boldsymbol{v}_2) = \alpha\lambda _1 \boldsymbol{v}_1 + \beta \lambda_2 \boldsymbol{v}_2$ . In the same way, $\boldsymbol{a}_2 = A\boldsymbol{a}_1 = A (\alpha\lambda _1 \boldsymbol{v}_1 + \beta \lambda_2 \boldsymbol{v}_2) = \alpha\lambda _{1}^{2} \boldsymbol{v}_1 + \beta \lambda_{2}^{2} \boldsymbol{v}_2$ , and $\boldsymbol{a}_3 = A\boldsymbol{a}_2 = A (\alpha\lambda _{1}^{2} \boldsymbol{v}_1 + \beta \lambda_{2}^{2} \boldsymbol{v}_2) = \alpha\lambda _{1}^{3} \boldsymbol{v}_1 + \beta \lambda_{2}^{3} \boldsymbol{v}_2$ . These equations show that in coordinate system made by eigen vectors of $A$ , linear transformation by $A$ is easily done by just multiplying eigen values with each eigen vector. Compared to the graph of Fibonacci numbers above, in the figure below you can see that in coordinate system made by eigen vectors the plots changes more systematically generation by generation.

In coordinate system made by eigen vectors of square matrices, the linear transformations by the matrices can be much more straightforward, and this is one powerful strength of eigen vectors.

*I do not major in mathematics, so I am not 100% sure, but vectors in linear algebra have more abstract meanings and various things in mathematics can be vectors, even though in machine learning or data science we mainly use numerical vectors with more concrete elements. We can also say that matrices are a kind of maps. That is just like, at leas in my impression, even though a real town is composed of various components such as houses, smooth or bumpy roads, you can simplify its structure with simple orthogonal lines, like the map of Manhattan. But if you know what the town actually looks like, you do not have to follow the zigzag path on the map.

4, Eigen vectors of positive semidefinite matrices

In the second section of this article I told you that, even though you have to consider various elements when you discuss general diagonalization, in terms of PCA and LDA we mainly use only a type of matrices named positive semidefinite matrices. Let $A$ be a $D \times D$ square matrix. If $\boldsymbol{x}^T A \boldsymbol{x} \geq 0$ for all values of the vector $\boldsymbol{x}$ , the $A$ is said to be a positive semidefinite matrix. And also it is known that $A$ being a semidefinite matrix is equivalent to $\lambda _{i} \geq 0$ for all the eigen values $\lambda_i (i=1, \dots , D)$ .

*I think most people first learn a type of matrices called positive definite matrices. Let $A$ be a $D \times D$ square matrix. If $\boldsymbol{x}^T A \boldsymbol{x} > 0$ for all values of the vector $\boldsymbol{x}$ , the $A$ is said to be a positive definite matrix. You have to keep it in mind that even if all the elements of $A$ are positive, $A$ is not necessarly positive definite/semidefinite.

Just as we did in the second section of this article, let’s visualize displacement vectors made by linear transformation with a $3 \times 3$ square positive semidefinite matrix $A$ .

*In fact $A_1 =\begin{pmatrix} 3 & 1 \\ 1 & 2 \end{pmatrix}$ , whose linear transformation I visualized the second section, is also positive semidefinite.

Let’s visualize linear transformations by a positive definite matrix $A = \frac{1}{50} \begin{pmatrix} 60.45 & 33.63 & 46.29 \\33.63 & 68.49 & 50.93 \\ 46.29 & 50.93 & 53.61 \end{pmatrix}$ . I visualized the displacement vectors made by the $A$ just as the same way as in the second section of this article. The result is as below, and you can see that, as well as the displacement vectors made by $A_1$ , the three dimensional displacement vectors below are swirling and extending in three directions, in the directions of the three orthogonal eigen vectors $\boldsymbol{v}_1, \boldsymbol{v}_2$ , and $\boldsymbol{v}_3$ .

*It might seem like a weird choice of a matrix, but you are going to see why in the next article.

You might have already noticed $A_1 =\begin{pmatrix} 3 & 1 \\ 1 & 2 \end{pmatrix}$ and $A = \frac{1}{50} \begin{pmatrix} 60.45 & 33.63 & 46.29 \\33.63 & 68.49 & 50.93 \\ 46.29 & 50.93 & 53.61 \end{pmatrix}$ are both symmetric matrices and that their elements are all real values, and that their diagonal elements are all positive values. Super importantly, when all the elements of a $D \times D$ symmetric matrix $A$ are real values and its eigen values are $\lambda_{i} (i=1, \dots , D)$ , there exist orthonormal matrices $U$ such that $U^{-1}AU = \Lambda$ , where $\Lambda = diag(\lambda_{1}, \dots , \lambda_{D})$ .

*The title of this section might be misleading, but please keep it in mind that positive definite/semidefinite matrices are not necessarily real symmetric matrices. And real symmetric vectors are not necessarily positive definite/semidefinite matrices.

5, Orthonormal matrices and rotation of vectors

In this section I am gong to explain orthonormal matrices, as known as rotation matrices. If a $D\times D$ matrix $U$ is an orthonormal matrix, column vectors of $U$ are orthonormal, which means $U = (\boldsymbol{u}_1 \dots \boldsymbol{u}_D)$ , where $\begin{cases} \boldsymbol{u}_{i}^{T}\boldsymbol{u}_{j} = 1 \quad (i = j) \\ \boldsymbol{u}_{i}^{T}\boldsymbol{u}_{j} = 0 \quad (i\neq j) \end{cases}$ . In other words column vectors $\boldsymbol{u}_{i}$ form an orthonormal coordinate system.

Orthonormal matrices $U$ have several important matrices, and one of them is $U^{-1} = U^{T}$ . Combining this fact with what I have told you so far, you we can reach one conclusion that you can orthogonalize a real symmetric matrix $A$ as $U^{T}AU = \Lambda$ . This is known as spectral decomposition or singular value decomposition.

Another important property of $U$ is that $U^{T}$ is also orthonormal. In other words, assume $U$ is orthonormal and that $U = (\boldsymbol{u}_1 \dots \boldsymbol{u}_D) = \begin{pmatrix} -\boldsymbol{v_1}^{T}- \\ \vdots \\ -\boldsymbol{v_D}^{T}- \end{pmatrix}$ , $(\boldsymbol{v}_1 \dots \boldsymbol{v}_D)$ also forms a orthonormal coordinate system.

…It seems things are getting too mathematical and abstract (for me), thus for now I am going to wrap up what I have explained in this article .

We have seen

Numerical matrices linearly transform vectors.
Certain linear transformations do not change the direction of vectors in certain directions, which are called eigen vectors.
Making use of eigen vectors, you can form new coordinate system which can describe the linear transformations in a more straightforward way.
You can diagonalize a real symmetric matrix $A$ with an orthonormal matrix $U$ .

Of our current interest is what kind of linear transformation the real symmetric positive definite matrix enables. I am going to explain why the purple vectors in the figure above is swirling like that in the upcoming articles. Before that, however, we are going to see one application of what we have seen in this article, on dimension reduction. To be concrete the next article is going to be about principal component analysis (PCA), which is very important in many fields.

*In short, the orthonormal matrix $U$ I mentioned above enables rotation of matrix, and the diagonal matrix $diag(\lambda_1, \dots, \lambda_D)$ expands or contracts vectors along each axis. I am going to explain that more precisely in the upcoming articles.

* I make study materials on machine learning, sponsored by DATANOMIQ. I do my best to make my content as straightforward but as precise as possible. I include all of my reference sources. If you notice any mistakes in my materials, including grammatical errors, please let me know (email: yasuto.tamura@datanomiq.de). And if you have any advice for making my materials more understandable to learners, I would appreciate hearing it.

*I attatched the codes I used to make the figures in this article. You can just copy, paste, and run, sometimes installing necessary libraries.

import matplotlib.pyplot as plt 
import numpy as np
import matplotlib.patches as mpatches

T_A = np.array([[1, 1], 
                [1, 0]])

total_step = 5
x = np.zeros((total_step, 2))

x[0]  = np.array([1, 0])

for i in range(total_step - 1):
    x[i + 1] = np.dot(T_A, x[i])

eigen_values, eigen_vectors = np.linalg.eig(T_A)
idx = eigen_values.argsort()[::-1]   
eigen_values = eigen_values[idx]
eigen_vectors = eigen_vectors[:,idx]
for i in range(len(eigen_vectors)):
    if(eigen_vectors.T[i][0] < 0):
        eigen_vectors.T[i] = - eigen_vectors.T[i]   

v_initial = x[0]
v_coefficients = np.zeros((total_step , 2))
v_coefficients[0] = np.dot(v_initial ,  np.linalg.inv(eigen_vectors.T)) 

for i in range(total_step-1):
    v_coefficients[i + 1] = v_coefficients[i] * eigen_values 

for i in range(total_step):
    v_1_list[i+1] = v_coefficients.T[0][i]*eigen_vectors.T[0]
    v_2_list[i+1] = v_coefficients.T[1][i]*eigen_vectors.T[1]

plt.figure(figsize=(20, 15))
fontsize = 20
small_shift = 0.2

plt.plot(x[:, 0], x[:, 1], marker='o', linestyle='none', markersize=10, color='black')

plt.arrow(0, 0, eigen_vectors.T[0][0], eigen_vectors.T[0][1], width=0.05, head_width=0.2, color='orange')
plt.arrow(0, 0, eigen_vectors.T[1][0], eigen_vectors.T[1][1], width=0.05, head_width=0.2, color='orange')

plt.text(eigen_vectors.T[0][0], eigen_vectors.T[0][1]+small_shift, r'', va='center',ha='right', fontsize=fontsize + 10)
plt.text(eigen_vectors.T[1][0] - small_shift, eigen_vectors.T[1][1],r'', va='center',ha='right', fontsize=fontsize + 10)

for i in range(total_step): 
    
    plt.arrow(0, 0, v_1_list[i+1][0], v_1_list[i+1][1], head_width=0.05, color='darkviolet', length_includes_head=True)
    plt.arrow(0, 0, v_2_list[i+1][0], v_2_list[i+1][1], head_width=0.05, color='darkviolet', length_includes_head=True)
    
    plt.text(v_1_list[i+1][0] + 2*small_shift , v_1_list[i+1][1]-2*small_shift,r''.format(1,i+1, 1),va='center',ha='right', fontsize=fontsize)
    plt.text(v_2_list[i+1][0]-0.1, v_2_list[i+1][1],r''.format(2, i+1, 2),va='center',ha='right', fontsize=fontsize)

    plt.arrow(v_1_list[i+1][0], v_1_list[i+1][1], v_2_list[i+1][0], v_2_list[i+1][1], head_width=0, color='black', linestyle='--', length_includes_head=True)
    plt.arrow(v_2_list[i+1][0], v_2_list[i+1][1], v_1_list[i+1][0], v_1_list[i+1][1], head_width=0, color='black', linestyle='--', length_includes_head=True)
    
orange_patch = mpatches.Patch(color='orange', label='Eigen vectors')
purple_patch = mpatches.Patch(color='darkviolet', label='Scalar multiples of the eigen vectors')
plt.legend(handles=[orange_patch, purple_patch], fontsize=25, loc='lower right')

for i in range(total_step):
    plt.text(x[i][0]+0.1, x[i][1]-0.05, r'n={0}'.format(i), fontsize=20)

plt.grid(True)
plt.ylabel("", fontsize=20)
plt.xlabel("", fontsize=20)
plt.title("Fibonacci sequence and its eigen space", fontsize=30)
#plt.savefig("Fibonacci_eigen_space.png")
plt.show()

import matplotlib.pyplot as plt 
import numpy as np 
import matplotlib.patches as mpatches

const_range = 10

X = np.arange(-const_range, const_range + 1, 1)
Y = np.arange(-const_range, const_range + 1, 1)
U_x, U_y = np.meshgrid(X, Y)

T_A_0 = np.array([[3, 1], 
                [1, 2]])

T_A_1 = np.array([[3, 1],
                [-1, 1]])

T_A_2 = np.array([[1, -1], 
                [1, 1]])

T_A_list = np.array((T_A_0, T_A_1, T_A_2))

const_range = 5
plt.figure(figsize=(30, 10))
plt.subplots_adjust(wspace=0.1)
labels = ["Grids", "Displacement vectors made by A", "Real eigen vectors of A"]  
title_list = [r" has two different real eigen vectors.", r" has two identical real unit eigen vectors.",  r" has only imaginary eigen vectors."]
for idx in range(len(T_A_list)): 
    
    eigen_values, eigen_vectors = np.linalg.eig(T_A_list[idx])
    sorted_idx = eigen_values.argsort()[::-1]   
    eigen_values = eigen_values[sorted_idx]
    eigen_vectors = eigen_vectors[:,sorted_idx]
    eigen_vectors = eigen_vectors.astype(float)
        
    for i in range(len(eigen_vectors)):
        if(eigen_vectors.T[i][0] < 0):
            eigen_vectors.T[i] = - eigen_vectors.T[i]
        

    X = np.arange(-const_range, const_range + 1, 1)
    Y = np.arange(-const_range, const_range + 1, 1)
    U_x, U_y = np.meshgrid(X, Y)

    V_x = np.zeros((len(U_x), len(U_y)))
    V_y = np.zeros((len(U_x), len(U_y)))

    temp_vec = np.zeros((1, 2))

    W_x = np.zeros((len(U_x), len(U_y)))
    W_y = np.zeros((len(U_x), len(U_y)))

    plt.subplot(1, 3, idx + 1)


    for i in range(len(U_x)):
        for j in range(len(U_y)):
            temp_vec[0][0] = U_x[i][j]
            temp_vec[0][1] = U_y[i][j]
        
            temp_vec[0] = np.dot(T_A_list[idx], temp_vec[0])
        
            V_x[i][j] = temp_vec[0][0]
            V_y[i][j] = temp_vec[0][1]
        
            W_x[i][j] = V_x[i][j] - U_x[i][j]
            W_y[i][j] = V_y[i][j] - U_y[i][j]
            #plt.arrow(0, 0, V_x[i][j], V_y[i][j], head_width=0.1, color='red')
            plt.arrow(0, 0, U_x[i][j], U_y[i][j], head_width=0.3, color='dimgrey', label=labels[0])
            plt.arrow(U_x[i][j], U_y[i][j], W_x[i][j], W_y[i][j], head_width=0.3, color='darkviolet', label=labels[1])
            
            range_const = 20
            plt.xlim([-range_const, range_const])
            plt.ylim([-range_const, range_const])
            plt.title(title_list[idx], fontsize=25)
            
            if idx==2:
                continue
 
            plt.arrow(0, 0, eigen_vectors.T[0][0]*10, eigen_vectors.T[0][1]*10, head_width=1, color='orange', label=labels[2])
            plt.arrow(0, 0, eigen_vectors.T[1][0]*10, eigen_vectors.T[1][1]*10, head_width=1, color='orange', label=labels[2])

grey_patch = mpatches.Patch(color='grey', label='Grids')
purple_patch = mpatches.Patch(color='darkviolet', label='Displacement vectors made by A')
yellow_patch = mpatches.Patch(color='gold', label='Real eigen vectors of A')
plt.legend(handles=[grey_patch, purple_patch, yellow_patch], fontsize=25, loc='lower right', bbox_to_anchor=(-0.1, -.35))
#plt.savefig("linear_transformation.png")
plt.show()

import numpy as np 
import matplotlib.pyplot as plt
from mpl_toolkits.mplot3d.proj3d import proj_transform
from mpl_toolkits.mplot3d.axes3d import Axes3D
from matplotlib.text import Annotation
from matplotlib.patches import FancyArrowPatch
import matplotlib.patches as mpatches

class Annotation3D(Annotation):
    def __init__(self, text, xyz, *args, **kwargs):
        super().__init__(text, xy=(0,0), *args, **kwargs)
        self._xyz = xyz

    def draw(self, renderer):
        x2, y2, z2 = proj_transform(*self._xyz, renderer.M)
        self.xy=(x2,y2)
        super().draw(renderer)

def _annotate3D(ax,text, xyz, *args, **kwargs):
    '''Add anotation `text` to an `Axes3d` instance.'''

    annotation= Annotation3D(text, xyz, *args, **kwargs)
    ax.add_artist(annotation)

setattr(Axes3D,'annotate3D',_annotate3D)

class Arrow3D(FancyArrowPatch):
    def __init__(self, x, y, z, dx, dy, dz, *args, **kwargs):
        super().__init__((0,0), (0,0), *args, **kwargs)
        self._xyz = (x,y,z)
        self._dxdydz = (dx,dy,dz)

    def draw(self, renderer):
        x1,y1,z1 = self._xyz
        dx,dy,dz = self._dxdydz
        x2,y2,z2 = (x1+dx,y1+dy,z1+dz)

        xs, ys, zs = proj_transform((x1,x2),(y1,y2),(z1,z2), renderer.M)
        self.set_positions((xs[0],ys[0]),(xs[1],ys[1]))
        super().draw(renderer)

def _arrow3D(ax, x, y, z, dx, dy, dz, *args, **kwargs):
    '''Add an 3d arrow to an `Axes3D` instance.'''

    arrow = Arrow3D(x, y, z, dx, dy, dz, *args, **kwargs)
    ax.add_artist(arrow)

setattr(Axes3D,'arrow3D',_arrow3D)

T_A = np.array([[60.45, 33.63, 46.29], 
                [33.63, 68.49, 50.93], 
                [46.29, 50.93, 53.61]])

T_A = T_A/50
const_range = 2


X = np.arange(-const_range, const_range + 1, 1)
Y = np.arange(-const_range, const_range + 1, 1)
Z = np.arange(-const_range, const_range + 1, 1)

U_x, U_y, U_z = np.meshgrid(X, Y, Z)

V_x = np.zeros((len(U_x), len(U_y), len(U_z)))
V_y = np.zeros((len(U_x), len(U_y), len(U_z)))
V_z = np.zeros((len(U_x), len(U_y), len(U_z)))

temp_vec = np.zeros((1, 3))

W_x = np.zeros((len(U_x), len(U_y), len(U_z)))
W_y = np.zeros((len(U_x), len(U_y), len(U_z)))
W_z = np.zeros((len(U_x), len(U_y), len(U_z)))

eigen_values, eigen_vectors = np.linalg.eig(T_A)
sorted_idx = eigen_values.argsort()[::-1]   
eigen_values = eigen_values[sorted_idx]
eigen_vectors = eigen_vectors[:,sorted_idx]
eigen_vectors = eigen_vectors.astype(float)

fig = plt.figure(figsize=(15, 15))
ax = fig.add_subplot(111, projection='3d')
grid_range = const_range + 5
ax.set_xlim(-grid_range, grid_range)
ax.set_ylim(-grid_range, grid_range)
ax.set_zlim(-grid_range, grid_range)

eigen_values, eigen_vectors = np.linalg.eig(T_A)
sorted_idx = eigen_values.argsort()[::-1]   
eigen_values = eigen_values[sorted_idx]
eigen_vectors = eigen_vectors[:,sorted_idx]
eigen_vectors = eigen_vectors.astype(float)
     
    
for i in range(len(eigen_vectors)):
    if(eigen_vectors.T[i][0] < 0):
        eigen_vectors.T[i] = - eigen_vectors.T[i]

for i in range(len(U_x)):
    for j in range(len(U_x)):
        for k in range(len(U_x)):
            temp_vec[0][0] = U_x[i][j][k]
            temp_vec[0][1] = U_y[i][j][k]
            temp_vec[0][2] = U_z[i][j][k]
        
            temp_vec[0] = np.dot(T_A, temp_vec[0])
        
            V_x[i][j][k] = temp_vec[0][0]
            V_y[i][j][k] = temp_vec[0][1]
            V_z[i][j][k] = temp_vec[0][2]
        
            W_x[i][j][k] = V_x[i][j][k] - U_x[i][j][k]
            W_y[i][j][k] = V_y[i][j][k] - U_y[i][j][k]
            W_z[i][j][k] = V_z[i][j][k] - U_z[i][j][k]
            ax.arrow3D(0, 0, 0, 
                       U_x[i][j][k], U_y[i][j][k], U_z[i][j][k], 
                       mutation_scale=10, arrowstyle="-|>", fc='dimgrey', ec='dimgrey')
            #ax.arrow3D(0, 0, 0, 
            #          V_x[i][j][k], V_y[i][j][k], V_z[i][j][k], 
            #           mutation_scale=10, arrowstyle="-|>", fc='red', ec='red')
            ax.arrow3D(U_x[i][j][k], U_y[i][j][k], U_z[i][j][k], 
                       W_x[i][j][k], W_y[i][j][k], W_z[i][j][k],
                       mutation_scale=10, arrowstyle="-|>", fc='darkviolet', ec='darkviolet')
            
ax.arrow3D(0, 0, 0, eigen_vectors.T[0][0]*10, eigen_vectors.T[0][1]*10, eigen_vectors.T[0][2]*10,
                       mutation_scale=10,  arrowstyle="-|>", fc='orange', ec='orange')
ax.arrow3D(0, 0, 0, eigen_vectors.T[1][0]*10, eigen_vectors.T[1][1]*10, eigen_vectors.T[1][2]*10,
                       mutation_scale=10, arrowstyle="-|>", fc='orange', ec='orange')
ax.arrow3D(0, 0, 0, eigen_vectors.T[2][0]*10, eigen_vectors.T[2][1]*10, eigen_vectors.T[2][2]*10,
                       mutation_scale=10, arrowstyle="-|>", fc='orange', ec='orange')

ax.text(eigen_vectors.T[0][0]*8 , eigen_vectors.T[0][1]*8, eigen_vectors.T[0][2]*8+1, r'', fontsize=20)
ax.text(eigen_vectors.T[1][0]*8 , eigen_vectors.T[1][1]*8, eigen_vectors.T[1][2]*8, r'', fontsize=20)
ax.text(eigen_vectors.T[2][0]*8 , eigen_vectors.T[2][1]*8, eigen_vectors.T[2][2]*8, r'', fontsize=20)


grey_patch = mpatches.Patch(color='grey', label='Grids')
orange_patch = mpatches.Patch(color='orange', label='Orthogonal eigen vectors of A')
purple_patch = mpatches.Patch(color='darkviolet', label='Displacement vectors made by A')
plt.legend(handles=[grey_patch, orange_patch, purple_patch], fontsize=20, loc='lower right')

ax.set_xlabel(r'', fontsize=25)
ax.set_ylabel(r'', fontsize=25)
ax.set_zlabel(r'', fontsize=25)
#plt.savefig("symmetric_positive_definite_visualizaiton.png")
plt.show()

Simple RNN: the first foothold for understanding LSTM

June 17, 2020/in Artificial Intelligence, Data Science, Deep Learning, Machine Learning, Main Category, Mathematics/by Yasuto Tamura

*In this article “Densely Connected Layers” is written as “DCL,” and “Convolutional Neural Network” as “CNN.”

In the last article, I mentioned “When it comes to the structure of RNN, many study materials try to avoid showing that RNNs are also connections of neurons, as well as DCL or CNN.” Even if you manage to understand DCL and CNN, you can be suddenly left behind once you try to understand RNN because it looks like a different field. In the second section of this article, I am going to provide a some helps for more abstract understandings of DCL/CNN , which you need when you read most other study materials.

My explanation on this simple RNN is based on a chapter in a textbook published by Massachusetts Institute of Technology, which is also recommended in some deep learning courses of Stanford University.

First of all, you should keep it in mind that simple RNN are not useful in many cases, mainly because of vanishing/exploding gradient problem, which I am going to explain in the next article. LSTM is one major type of RNN used for tackling those problems. But without clear understanding forward/back propagation of RNN, I think many people would get stuck when they try to understand how LSTM works, especially during its back propagation stage. If you have tried climbing the mountain of understanding LSTM, but found yourself having to retreat back to the foot, I suggest that you read through this article on simple RNNs. It should help you to gain a solid foothold, and you would be ready for trying to climb the mountain again.

*This article is the second article of “A gentle introduction to the tiresome part of understanding RNN.”

1, A brief review on back propagation of DCL.

Simple RNNs are straightforward applications of DCL, but if you do not even have any ideas on DCL forward/back propagation, you will not be able to understand this article. If you more or less understand how back propagation of DCL works, you can skip this first section.

Deep learning is a part of machine learning. And most importantly, whether it is classical machine learning or deep learning, adjusting parameters is what machine learning is all about. Parameters mean elements of functions except for variants. For example when you get a very simple function $f(x)=a + bx + cx^2 + dx^3$ , then $x$ is a variant, and $a, b, c, d$ are parameters. In case of classical machine learning algorithms, the number of those parameters are very limited because they were originally designed manually. Such functions for classical machine learning is useful for features found by humans, after trial and errors(feature engineering is a field of finding such effective features, manually). You adjust those parameters based on how different the outputs(estimated outcome of classification/regression) are from supervising vectors(the data prepared to show ideal answers).

In the last article I said neural networks are just mappings, whose inputs are vectors, matrices, or sequence data. In case of DCLs, inputs are vectors. Then what’s the number of parameters ? The answer depends on the the number of neurons and layers. In the example of DCL at the right side, the number of the connections of the neurons is the number of parameters(Would you like to try to count them? At least I would say “No.”). Unlike classical machine learning you no longer need to do feature engineering, but instead you need to design networks effective for each task and adjust a lot of parameters.

*I think the hype of AI comes from the fact that neural networks find features automatically. But the reality is difficulty of feature engineering was just replaced by difficulty of designing proper neural networks.

It is easy to imagine that you need an efficient way to adjust those parameters, and the method is called back propagation (or just backprop). As long as it is about DCL backprop, you can find a lot of well-made study materials on that, so I am not going to cover that topic precisely in this article series. Simply putting, during back propagation, in order to adjust parameters of a layer you need errors in the next layer. And in order calculate the errors of the next layer, you need errors in the next next layer.

*You should not think too much about what the “errors” exactly mean. Such “errors” are defined in this context, and you will see why you need them if you actually write down all the mathematical equations behind backprops of DCL.

The red arrows in the figure shows how errors of all the neurons in a layer propagate backward to a neuron in last layer. The figure shows only some sets of such errors propagating backward, but in practice you have to think about all the combinations of such red arrows in the whole back propagation(this link would give you some ideas on how DCLs work).

These points are minimum prerequisites for continuing reading this RNN this article. But if you are planning to understand RNN forward/back propagation at an abstract/mathematical level that you can read academic papers, I highly recommend you to actually write down all the equations of DCL backprop. And if possible you should try to implement backprop of three-layer DCL.

2, Forward propagation of simple RNN

In fact the simple RNN which we are going to look at in this article has only three layers. From now on imagine that inputs of RNN come from the bottom and outputs go up. But RNNs have to keep information of earlier times steps during upcoming several time steps because as I mentioned in the last article RNNs are used for sequence data, the order of whose elements is important. In order to do that, information of the neurons in the middle layer of RNN propagate forward to the middle layer itself. Therefore in one time step of forward propagation of RNN, the input at the time step propagates forward as normal DCL, and the RNN gives out an output at the time step. And information of one neuron in the middle layer propagate forward to the other neurons like yellow arrows in the figure. And the information in the next neuron propagate forward to the other neurons, and this process is repeated. This is called recurrent connections of RNN.

*To be exact we are just looking at a type of recurrent connections. For example Elman RNNs have simpler recurrent connections. And recurrent connections of LSTM are more complicated.

Whether it is a simple one or not, basically RNN repeats this process of getting an input at every time step, giving out an output, and making recurrent connections to the RNN itself. But you need to keep the values of activated neurons at every time step, so virtually you need to consider the same RNNs duplicated for several time steps like the figure below. This is the idea of unfolding RNN. Depending on contexts, the whole unfolded DCLs with recurrent connections is also called an RNN.

In many situations, RNNs are simplified as below. If you have read through this article until this point, I bet you gained some better understanding of RNNs, so you should little by little get used to this more abstract, blackboxed way of showing RNN.

You have seen that you can unfold an RNN, per time step. From now on I am going to show the simple RNN in a simpler way, based on the MIT textbook which I recomment. The figure below shows how RNN propagate forward during two time steps $(t-1), (t)$ .

The input $\boldsymbol{x}^{(t-1)}$ at time step $(t-1)$ propagate forward as a normal DCL, and gives out the output $\hat{\boldsymbol{y}} ^{(t)}$ (The notation on the $\boldsymbol{y} ^{(t)}$ is called “hat,” and it means that the value is an estimated value. Whatever machine learning tasks you work on, the outputs of the functions are just estimations of ideal outcomes. You need to adjust parameters for better estimations. You should always be careful whether it is an actual value or an estimated value in the context of machine learning or statistics). But the most important parts are the middle layers.

*To be exact I should have drawn the middle layers as connections of two layers of neurons like the figure at the right side. But I made my figure closer to the chart in the MIT textbook, and also most other study materials show the combinations of the two neurons before/after activation as one neuron.

$\boldsymbol{a}^{(t)}$ is just linear summations of $\boldsymbol{x}^{(t)}$ (If you do not know what “linear summations” mean, please scroll this page a bit), and $\boldsymbol{h}^{(t)}$ is a combination of activated values of $\boldsymbol{a}^{(t)}$ and linear summations of $\boldsymbol{h}^{(t-1)}$ from the last time step, with recurrent connections. The values of $\boldsymbol{h}^{(t)}$ propagate forward in two ways. One is normal DCL forward propagation to $\hat{\boldsymbol{y}} ^{(t)}$ and $\boldsymbol{o}^{(t)}$ , and the other is recurrent connections to $\boldsymbol{h}^{(t+1)}$ .

These are equations for each step of forward propagation.

$\boldsymbol{a}^{(t)} = \boldsymbol{b} + \boldsymbol{W} \cdot \boldsymbol{h}^{(t-1)} + \boldsymbol{U} \cdot \boldsymbol{x}^{(t)}$
$\boldsymbol{h}^{(t)}= g(\boldsymbol{a}^{(t)})$
$\boldsymbol{o}^{(t)} = \boldsymbol{c} + \boldsymbol{V} \cdot \boldsymbol{h}^{(t)}$
$\hat{\boldsymbol{y}} ^{(t)} = f(\boldsymbol{o}^{(t)})$

*Please forgive me for adding some mathematical equations on this article even though I pledged not to in the first article. You can skip the them, but for some people it is on the contrary more confusing if there are no equations. In case you are allergic to mathematics, I prescribed some treatments below.

*Linear summation is a type of weighted summation of some elements. Concretely, when you have a vector $\boldsymbol{x}=(x_0, x_1, x_2)$ , and weights $\boldsymbol{w}=(w_0,w_1, w_2)$ , then $\boldsymbol{w}^T \cdot \boldsymbol{x} = w_0 \cdot x_0 + w_1 \cdot x_1 +w_2 \cdot x_2$ is a linear summation of $\boldsymbol{x}$ , and its weights are $\boldsymbol{w}$ .

*When you see a product of a matrix and a vector, for example a product of $\boldsymbol{W}$ and $\boldsymbol{v}$ , you should clearly make an image of connections between two layers of a neural network. You can also say each element of $\boldsymbol{u}}$ is a linear summations all the elements of $\boldsymbol{v}}$ , and $\boldsymbol{W}$ gives the weights for the summations.

A very important point is that you share the same parameters, in this case $\boldsymbol{\theta \in \{\boldsymbol{U}, \boldsymbol{W}, \boldsymbol{b}, \boldsymbol{V}, \boldsymbol{c} \}}$ , at every time step.

And you are likely to see this RNN in this blackboxed form.

3, The steps of back propagation of simple RNN

In the last article, I said “I have to say backprop of RNN, especially LSTM (a useful and mainstream type or RNN), is a monster of chain rules.” I did my best to make my PowerPoint on LSTM backprop straightforward. But looking at it again, the LSTM backprop part still looks like an electronic circuit, and it requires some patience from you to understand it. If you want to understand LSTM at a more mathematical level, understanding the flow of simple RNN backprop is indispensable, so I would like you to be patient while understanding this step (and you have to be even more patient while understanding LSTM backprop).

This might be a matter of my literacy, but explanations on RNN backprop are very frustrating for me in the points below.

Most explanations just show how to calculate gradients at each time step.
Most study materials are visually very poor.
Most explanations just emphasize that “errors are back propagating through time,” using tons of arrows, but they lack concrete instructions on how actually you renew parameters with those errors.

If you can relate to the feelings I mentioned above, the instructions from now on could somewhat help you. And I am going to share some study materials on simple RNNs in an external link so that you can gain a clear and mathematical understanding on how simple RNNs work.

Backprop of RNN , as long as you are thinking about simple RNNs, is not so different from that of DCLs. But you have to be careful about the meaning of errors in the context of RNN backprop. Back propagation through time (BPTT) is one of the major methods for RNN backprop, and I am sure most textbooks explain BPTT. But most study materials just emphasize that you need errors from all the time steps, and I think that is very misleading and confusing.

You need all the gradients to adjust parameters, but you do not necessarily need all the errors to calculate those gradients. Gradients in the context of machine learning mean partial derivatives of error functions (in this case $J$ ) with respect to certain parameters, and mathematically a gradient of $J$ with respect to $\boldsymbol{\theta \in \{\boldsymbol{U}, \boldsymbol{W}, \boldsymbol{b}^{(t)}, \boldsymbol{V}, \boldsymbol{c} \}}$ is denoted as $( \frac{\partial J}{\partial \boldsymbol{\theta}} )$ . And another confusing point in many textbooks, including the MIT one, is that they give an impression that parameters depend on time steps. For example some study materials use notations like $\frac{\partial J}{\partial \boldsymbol{\theta}^{(t)}}$ , and I think this gives an impression that this is a gradient with respect to the parameters at time step $(t)$ . In my opinion this gradient rather should be written as $( \frac{\partial J}{\partial \boldsymbol{\theta}} )^{(t)}$ . But many study materials denote gradients of those errors in the former way, so from now on let me use the notations which you can see in the figures in this article.

In order to calculate the gradient $\frac{\partial J}{\partial \boldsymbol{x}^{(t)}}$ you need errors from time steps $s (s \geq t) \quad$ (as you can see in the figure, in order to calculate a gradient in a colored frame, you need all the errors in the same color).

*Another confusing point is that the $\frac{\partial J}{\partial \boldsymbol{\ast ^{(t)}}}, \boldsymbol{\ast} \in \{\boldsymbol{a}^{(t)}, \boldsymbol{h}^{(t)}, \boldsymbol{o}^{(t)}, \dots \}$ are correct notations, because $\boldsymbol{\ast}$ are values of neurons after forward propagation. They depend on time steps, and these are very values which I have been calling “errors.” That is why parameters do not depend on time steps, whereas errors depend on time steps.

As I mentioned before, you share the same parameters at every time step. Again, please do not assume that parameters are different from time step to time step. It is gradients/errors (you need errors to calculate gradients) which depend on time step. And after calculating errors at every time step, you can finally adjust parameters one time, and that’s why this is called “back propagation through time.” (It is easy to imagine that this method can be very inefficient. If the input is the whole text on a Wikipedia link, you need to input all the sentences in the Wikipedia text to renew parameters one time. To solve this problem there is a backprop method named “truncated BPTT,” with which you renew parameters based on a part of a text. )

And after calculating those gradients $\frac{\partial J}{\partial \boldsymbol{\theta}^{(t)}}$ you can take a summation of them: $\frac{\partial J}{\partial \boldsymbol{\theta}}=\sum_{t=0}^{t=\tau}{\frac{\partial J}{\partial \boldsymbol{\theta}^{(t)}}}$ . With this gradient $\frac{\partial J}{\partial \boldsymbol{\theta}}$ , you can finally renew the value of $\boldsymbol{\theta}$ one time.

At the beginning of this article I mentioned that simple RNNs are no longer for practical uses, and that comes from exploding/vanishing problem of RNN. This problem was one of the reasons for the AI winter which lasted for some 20 years. In the next article I am going to write about LSTM, a fancier type of RNN, in the context of a history of neural network history.