POLICY GRADIENTS IN DEEP REINFORCEMENT LEARNING

In 2016, a deep learning Reinforcement agent AlphaGobeat Lee Sedol, who is a professional Go player of 9 dan rank (the highest honours in the field of Go). Go was regarded as a game far-fetched from computer algorithms because it has an artistic essence to it. But AlphaGo, developed by Google DeepMind was able to beat Lee Sedol 4–1 in a 5 match series. This event really sent a shrill to human intelligence all over the world, it becomes apparent from the fact that Lee Sedol took a cigarette break during one of the matches.

Reinforcement learning, unlike Supervised learning, works based on a reward function compared to data labels. In Reinforcement Learning, we have state, actions and rewards. The agent has to come up with a strategy to maximize its rewards. An RL agent has two components, i) description of the state based on Value functions, ii) policy distribution. The first one has algorithms like Q-learning, DQN, DDQN etc. The second one has different algorithms which we are going to discuss in this article.

LIMITATIONS OF VALUE-BASED METHODS

• The policy is simply mapping a state to different possible actions. So given our state, which action to take so that going forward we can maximize our reward.
• In the value-function method, first, we randomly choose a policy and you are dropped at a random state, then following that policy we estimate how good are the states that we are visiting. The states that give higher rewards going forward have a higher value function. So while building your final policy, you will choose states greedily (also you need to explore sometimes by choosing a state randomly by introducing some noise to our pre-defined randomized policy while learning).
• But sometimes it is better to learn a policy than accessing the quality of the state. Drawing parallels to real-world, given your present conditions, you are better off if you know what to do rather than how good or bad your present condition is.
• Also, value-based methods are not feasible in continuous action space. Here given a state, we need to pick an action that maximizes our action-value function. Algorithms like Q-learning are great with higher observational state spaces like Atari games which have 80x80 (image dimensions) state space, but they cannot handle higher dimensional action spaces. In the paper DDPG, the authors give an example of the Human arm which has 7 degrees of freedom and even if we will consider 4 possible actions (up, down, left and right) in each DOF, we will 4⁷ = 16384 possible action space. It is difficult to explore all these actions with techniques like epsilon greedy.
• Q-learning methods work by minimizing the TD error between our estimate of reward along with the action-value function of NEXT STATE (Sₜ₊₁), and the action-value function of the present state. But in the continuous domain, NEXT STATE (Sₜ)needs proper discretization so that we are not throwing away a lot of information and it is hard to achieve.

Q-learning update

• So to ameliorate these problems, we should try to learn the policy of an agent rather than only considering the value function. PG methods have their own set of problems, so a new set of Algorithms called Actor-Critic are developed which consider both policy and value functions. But here we will concentrate on PG methods, particularly REINFORCE.

#machine-learning #data-science #artificial-intelligence

What is GEEK

Buddha Community

Top Deep Learning Development Services | Hire Deep Learning Developer

View more: https://www.inexture.com/services/deep-learning-development/

We at Inexture, strategically work on every project we are associated with. We propose a robust set of AI, ML, and DL consulting services. Our virtuoso team of data scientists and developers meticulously work on every project and add a personalized touch to it. Because we keep our clientele aware of everything being done associated with their project so there’s a sense of transparency being maintained. Leverage our services for your next AI project for end-to-end optimum services.

#deep learning development #deep learning framework #deep learning expert #deep learning ai #deep learning services

Attend The Full Day Hands-On Workshop On Reinforcement Learning

The Association of Data Scientists (AdaSci), a global professional body of data science and ML practitioners, is holding a full-day workshop on building games using reinforcement learning on Saturday, February 20.

Artificial intelligence systems are outperforming humans at many tasks, starting from driving cars, recognising images and objects, generating voices to imitating art, predicting weather, playing chess etc. AlphaGo, DOTA2, StarCraft II etc are a study in reinforcement learning.

Reinforcement learning enables the agent to learn and perform a task under uncertainty in a complex environment. The machine learning paradigm is currently applied to various fields like robotics, pattern recognition, personalised medical treatment, drug discovery, speech recognition, and more.

With an increase in the exciting applications of reinforcement learning across the industries, the demand for RL experts has soared. Taking the cue, the Association of Data Scientists, in collaboration with Analytics India Magazine, is bringing an extensive workshop on reinforcement learning aimed at developers and machine learning practitioners.

#ai workshops #deep reinforcement learning workshop #future of deep reinforcement learning #reinforcement learning #workshop on a saturday #workshop on deep reinforcement learning

Deep Reinforcement Learning for Video Games Made Easy

In this post, we will investigate how easily we can train a Deep Q-Network (DQN) agent (Mnih et al., 2015) for Atari 2600 games using the Google reinforcement learning library Dopamine. While many RL libraries exist, this library is specifically designed with four essential features in mind:

• Easy experimentation
• Flexible development
• Compact and reliable
• Reproducible

_We believe these principles makes __Dopamine _one of the best RL learning environment available today. Additionally, we even got the library to work on Windows, which we think is quite a feat!

In my view, the visualization of any trained RL agent is an absolute must in reinforcement learning! Therefore, we will (of course) include this for our own trained agent at the very end!

We will go through all the pieces of code required (which is** minimal compared to other libraries**), but you can also find all scripts needed in the following Github repo.

1. Brief Introduction to Reinforcement Learning and Deep Q-Learning

The general premise of deep reinforcement learning is to

“derive efficient representations of the environment from high-dimensional sensory inputs, and use these to generalize past experience to new situations.”

• Mnih et al. (2015)

As stated earlier, we will implement the DQN model by Deepmind, which only uses raw pixels and game score as input. The raw pixels are processed using convolutional neural networks similar to image classification. The primary difference lies in the objective function, which for the DQN agent is called the optimal action-value function

where_ rₜ is the maximum sum of rewards at time t discounted by γ, obtained using a behavior policy π = P(a_∣_s)_ for each observation-action pair.

There are relatively many details to Deep Q-Learning, such as Experience Replay (Lin, 1993) and an _iterative update rule. _Thus, we refer the reader to the original paper for an excellent walk-through of the mathematical details.

One key benefit of DQN compared to previous approaches at the time (2015) was the ability to outperform existing methods for Atari 2600 games using the same set of hyperparameters and only pixel values and game score as input, clearly a tremendous achievement.

2. Installation

This post does not include instructions for installing Tensorflow, but we do want to stress that you can use both the CPU and GPU versions.

Nevertheless, assuming you are using Python 3.7.x, these are the libraries you need to install (which can all be installed via pip):

tensorflow-gpu=1.15   (or tensorflow==1.15  for CPU version)
cmake
dopamine-rl
atari-py
matplotlib
pygame
seaborn
pandas

#reinforcement-learning #q-learning #games #machine-learning #deep-learning #deep learning

Top 10 Deep Learning Sessions To Look Forward To At DVDC 2020

The Deep Learning DevCon 2020, DLDC 2020, has exciting talks and sessions around the latest developments in the field of deep learning, that will not only be interesting for professionals of this field but also for the enthusiasts who are willing to make a career in the field of deep learning. The two-day conference scheduled for 29th and 30th October will host paper presentations, tech talks, workshops that will uncover some interesting developments as well as the latest research and advancement of this area. Further to this, with deep learning gaining massive traction, this conference will highlight some fascinating use cases across the world.

Here are ten interesting talks and sessions of DLDC 2020 that one should definitely attend:

Also Read: Why Deep Learning DevCon Comes At The Right Time

---

Adversarial Robustness in Deep Learning

By Dipanjan Sarkar

**About: **Adversarial Robustness in Deep Learning is a session presented by Dipanjan Sarkar, a Data Science Lead at Applied Materials, as well as a Google Developer Expert in Machine Learning. In this session, he will focus on the adversarial robustness in the field of deep learning, where he talks about its importance, different types of adversarial attacks, and will showcase some ways to train the neural networks with adversarial realisation. Considering abstract deep learning has brought us tremendous achievements in the fields of computer vision and natural language processing, this talk will be really interesting for people working in this area. With this session, the attendees will have a comprehensive understanding of adversarial perturbations in the field of deep learning and ways to deal with them with common recipes.

Read an interview with Dipanjan Sarkar.

Imbalance Handling with Combination of Deep Variational Autoencoder and NEATER

By Divye Singh

**About: **Imbalance Handling with Combination of Deep Variational Autoencoder and NEATER is a paper presentation by Divye Singh, who has a masters in technology degree in Mathematical Modeling and Simulation and has the interest to research in the field of artificial intelligence, learning-based systems, machine learning, etc. In this paper presentation, he will talk about the common problem of class imbalance in medical diagnosis and anomaly detection, and how the problem can be solved with a deep learning framework. The talk focuses on the paper, where he has proposed a synergistic over-sampling method generating informative synthetic minority class data by filtering the noise from the over-sampled examples. Further, he will also showcase the experimental results on several real-life imbalanced datasets to prove the effectiveness of the proposed method for binary classification problems.

Default Rate Prediction Models for Self-Employment in Korea using Ridge, Random Forest & Deep Neural Network

By Dongsuk Hong

About: This is a paper presentation given by Dongsuk Hong, who is a PhD in Computer Science, and works in the big data centre of Korea Credit Information Services. This talk will introduce the attendees with machine learning and deep learning models for predicting self-employment default rates using credit information. He will talk about the study, where the DNN model is implemented for two purposes — a sub-model for the selection of credit information variables; and works for cascading to the final model that predicts default rates. Hong’s main research area is data analysis of credit information, where she is particularly interested in evaluating the performance of prediction models based on machine learning and deep learning. This talk will be interesting for the deep learning practitioners who are willing to make a career in this field.

---

#opinions #attend dldc 2020 #deep learning #deep learning sessions #deep learning talks #dldc 2020 #top deep learning sessions at dldc 2020 #top deep learning talks at dldc 2020

Why there is a problem with the Policy Gradient theorem in Deep Reinforcement Learning

I first discovered this problem thanks to my supervisor, Prof. Proutiere, while checking my proposal for the second lab session of the Reinforcement Learning (EL2805) course we held at KTH last year, in fall 2020.

The problem is severe enough to affect most Deep Reinforcement Learning algorithms, including A3C [7], SAC [8], the ACKTR algorithm [9], and others.

Roughly, that’s what my supervisor emailed me:

Hey Alessio, I have a question regarding the policy gradient theorem. The average is with respect to μ, which is a discounted state distribution. But, how do generally people justify the fact that we are using experiences generated by the stationary on-policy distribution induced by the policy π, instead of the discounted distribution μ?

A couple of emails back and forth between my supervisor and me were necessary to fully understand what he meant (in my defense, his first email was left ambiguous, and the previous quoted sentence is a “clarified” version of that email :)).

If you, yet, do not know what I’m talking about, let me briefly remind you how the Policy Gradient (PG) theorem [1] works.

The Policy Gradient theorem

The PG theorem, despite its simplicity, has enabled many of the achievements in deep reinforcement learning that we are currently seeing.

#deep-learning #reinforcement-learning #deep-dives #machine-learning