Machine Learning

Hyperparameteroptimizationisbothapracticalissueandaninterestingtheoreticalproblemintraining of deep architectures. Despite many recent advances the most commonly used methods almost universally involve training multiple and decoupled copies of the model, in effect sampling the hyperparameter space. We show that at negligable additional computational cost, results can be improved by sampling paths instead of points in hyperparameter space. To this end we interpret hyperparameters as controlling the level of correlated noise in the training, which can be...

We propose a DQN based technique to generalize the concept of skills. In previous papers, we have seen attempts to hand-craft skills (Chen Tessler, 2016). The idea of skills is that skills reduce the effective size of the MDP by enabling the agent to plan at a higher level. More than that, every specific task will be performed in the best way. However, skills are specific for every domain. In...

Automatically generating images according to natural language descriptions is a fundamental problem in many applications such as computer-aided design, photo editing, art generation, and video games. AttnGAN is a neural architecture that combines the attention mechanism from the field of natural language processing, and the generative adversarial networks from the field of image processing. In this project, we implemented the architecture, suggested by the original researchers, trained the network on...

In this project we implemented a trading agent for cryptographic currencies, which uses the currencies price and volume history, as well as published posts and comments on Reddit forums. Its target is to change the portfolio appropriately, in order to maximize the profit. In the project we rely on the assumption that a correlation exists between the currency’s price and the textual publications about it on the different forums. To...

We address the problem of solving high dimensional MDPs using its minimization by sampling and solving a navigation problem on a graph using RMAX. Many previous attempts optimize the policy using deep models with some success but lack the theoretical guaranties. Our methods provide a framework to incorporate classical reinforcement learning algorithms which provide guaranties. Moreover, the method can be extended using deep models to solve more specific problems. The...

In our Project we created a system that can independently download movies reviews, classify them and decide for each review if it is positive or negative. The process of getting the reviews is done by an independent script (crawler) that can search in the internet and download a large amount of reviews. For classifying each review, we created a trained module based on machine learning algorithms with different kinds of...

We consider a distributed reinforcement learning framework where multiple agents interact with the environment in parallel, while sharing experience, in order to find the optimal policy. At each time step, only a subset of the agents is selected to interact with the environment. We explore several mechanisms for selecting which agents to prioritize based on the reward and the TD-error, and analyze their effect on the learning process. When the...

NeuroEvolution is a field in AI which uses evolutionary algorithms and can be used to solve reinforcement learning problems. The process in those algorithms is initializing simple neural net (usually on hidden neuron only) and then performing evolution on the weights space to get the best fitness score. This project is based on a paper [1] that shows that evolution of the topology can improve the performance in control tasks...

Project’s main objective was to develop a methodology for solving planning problems involving time by using simple planning tools. During the project I studied the field of classical and temporal planning, learned the PDDL language which is used for describing planning problems and designed a method for solving temporal planning problems using classical planning solver written in python (Pyperplan).

In this project, we consider the Inverse Reinforcement Learning problem in Contextual Markov Decision Processes. Here, the reward of the environment depends on a hidden static parameter referred to as the context, i.e., each context defines an MDP. The agent does not observe the reward, but instead, is provided with expert demonstrations for different contexts. The goal of the agent is to learn a mapping from contexts to rewards that...

Background: Deep reinforcement learning (DRL) methods such as the Deep Q-Network (DQN) have achieved state-of-the-art results in a variety of challenging, high-dimensional domains. One of the major problems in DRL is that the state and action spaces are large, and the task often becomes an exploration problem. While DRL methods are new, RL is a well established field. There exist various algorithms in the tabular case which ensure proper exploration and convergence to the...

Deep learning in general, and segmentation problems in particular, are the subject of much research done in the field of learning systems. The rising power of computers enables the implementation of complex networks containing thousands of variables in order to create optimal architecture for such problems. However, there are several approaches to dealing with segmentation problems, each of which has its own advantages and disadvantages. This project deals with a...

Extreme classification is a growing research area dealing with multi-class tasks with an extremely large number of classes. One new challenge arising in this setting is to perform inference in a reasonable time, namely quicker than common-practice multi-class algorithms which require linear time in the number of classes. Many recent studies in this area learn hierarchical classification models, which allow logarithmic inference time. In the project we propose a different...

We present a novel and effective approach for video to video translation through generative adversarial networks. Given an input video of a person and a target video, our model generates a video of that same person mimicking the motion of the person in the target video. Our model can be used for various tasks, from ”teaching” someone to dance like a skilled dancer to the generation of realistic videos of...

GAN – AI algorithm implemented by a system of two neural networks contesting each other in a zero sum game. Our goal was generating cars photos which don’t exist which look real to human by using a GAN.