## Reinforcement learning: policy gradient methods

Policy gradient methods are a type of Reinforcement Learning optimization methods that works by performing gradient ascent on the parameters of a parameterized policy. This

## NLP: what is attention mechanism?

In 2022, the NLP (natural language processing) benchmarks have been dominated by transformer models, and the attention mechanism is one of the key ingredients to

## On-policy Control with Approximate Value Functions

This is a continuation fromÂ Approximate Function Methods in Reinforcement learning Episodic Sarsa with Function Approximation Reminder of what Sarsa is State, Action, Reward, State, Action

## Approximate Function Methods in Reinforcement learning

Tabular vs Function Methods In reinforcement learning, there are a few methods that are called tabular methods because they track a table of the (input,

## Planning with Tabular Methods in Reinforcement Learning

Tabular methods Tabular methods refer to problems in which the state and actions spaces are small enough for approximate value functions to be represented as

## Temporal Difference Control in Reinforcement Learning

Temporal Difference learning is one of the most important idea in Reinforcement Learning. We should go over the control aspect of TD to find an

## Temporal Difference Learning

Temporal Difference (TD) learning is the most novel and central idea of reinforcement learning. It combines the advantages from Dynamic Programming and Monte Carlo methods.

## Monte Carlo Methods in RL

In Reinforcement Learning, the Monte Carlo methods are a collection of methods for estimating the value functions and discovering optimal policies thru experience – sampling

## Solving MDPs with dynamic programming

In Reinforcement Learning, one way to solve finite MDPs is to use dynamic programming. Policy Evaluation (of the value functions) It refers to the iterative

## Finite Markov Decision Processes (MDP)

Finite MDP is the formal problem definition that we try to solve in most of the reinforcement learning problem. Definition Finite MDP is a classical