Ryan Lee

Ryan Lee’s notes on AI agents, LLMs, web extraction, and practical automation.
A periwinkle brain facing its reflection in a mustard-framed mirror.

Reality check on self-distillation and Redesigning source control

llm
agents
agents-in-practice

Agents in Practice #16 on verifying the effectiveness of self-distillation against GRPO and Anthropic’s account on redesigning source control in the age of coding agents.

Oct 23, 2026
Ryan Lee
A blue brain inside a cream desktop computer with an orange power button.

LLMs building Codebases from Scratch and Measuring Agents in Production

llm
agents
agents-in-practice

Agents in Practice #15 on a new benchmark for evaluating coding agents building codebases from scratch and a survey on how teams build and evaluate agents deployed in production.

Oct 16, 2026
Ryan Lee
Blue and mustard annotation circles overlap around five blue stones, with a pencil beside each circle.

Cognitive Chain-of-Thought and How tools shape agent behavior

llm
agents
agents-in-practice

Agents in Practice #14 on Structured reasoning in Chain-of-Thought, and how tool interfaces can change agent behavior.

Oct 8, 2026
Ryan Lee
A mustard-colored eraser erases the last footprint in a staggered trail of blue shoe prints.

Agents in Practice #13: Agents tampering with traces and overclaiming their work

llm
agents
agents-in-practice

Agents in Practice #13 on agents tampering with traces and overclaiming their work.

Oct 1, 2026
Ryan Lee
A yellow whistle stops a chain of falling blue dominoes

Agents in Practice #12: Reacting to exploits and evaluating compression of multi-agent systems

llm
agents
agents-in-practice

Agents in Practice #12 on how agents react to exploits and the cautionary tale of evaluating compression of multi-agent systems.

Sep 24, 2026
Ryan Lee
An open book with a blue puzzle-shaped space and a mustard puzzle piece rising above it.

Agents in Practice #11: Evolving skills with a wiki and reconstructing environments from trajectories

llm
agents
agents-in-practice

Agents in Practice #11 on evolving skills with traces and a wiki and reconstructing environments from trajectories to fine-tune agents.

Sep 17, 2026
Ryan Lee
A precariously leaning stack of blue blocks with an orange block near the top

Agents in Practice #10: Fragile self-improving agents, and measuring what benchmarks measure

llm
agents
agents-in-practice

Agents in Practice #10 on the fragility of memory-based self-improving agents, and on a modeling method for determining whether benchmarks measure general reasoning or safety.

Sep 10, 2026
Ryan Lee
Magnifying glass revealing a dotted watermark pattern in a stream of blue tokens

Agents in Practice #9: Watermarking LLM output and Learning from Copilot traces

llm
agents
agents-in-practice

Agents in Practice #9 on Claude’s new policy of adding watermarks to LLM output, and Microsoft’s takeaways from analyzing Copilot traces

Sep 3, 2026
Ryan Lee
Preview graphic for Agents in Practice #8

Agents in Practice #8: Reproducing ICML 2026 papers and evaluating agent traces

llm
agents
agents-in-practice

Agents in Practice #8 on Hugging Face’s hackathon on reproducing ICML 2026 papers and a new benchmark evaluating LLM judges for agent traces.

Aug 20, 2026
Ryan Lee
Preview graphic for Agents in Practice #7

Agents in Practice #7: Agents breaking rules and agents using a desktop GUI

llm
agents
agents-in-practice

Agents in Practice #7 on recent cybersecurity incidents by Anthropic and OpenAI, and a new benchmark testing an agent’s ability to use computers

Aug 6, 2026
Ryan Lee
Preview graphic for Agents in Practice #6

Agents in Practice #6: Patient Agents and Databricks’ Coding Benchmark

llm
agents
agents-in-practice

Agents in Practice #6 on a benchmark that tests an agent’s ability to wait and detect changes, and the Databricks team’s findings from its internal coding benchmark

Jul 30, 2026
Ryan Lee
Three progressively smaller blue brain forms compressing into an orange module above a compact local computer

Agents in Practice #5: Proactive Memory Agent, Unreliable Benchmarks, and the Rise of Local LLMs

llm
agents
agents-in-practice

Agents in Practice #5 on a memory agent that learns when to intervene, audits uncovering flaws in coding benchmarks, and a personal anecdote about increasingly capable local LLMs.

Jul 23, 2026
Ryan Lee
Open research notebook with a single connected spiral line representing self-referencing agent research logs

Agents in Practice #4: Coding Agent Latents, Test-Writing Benchmark, and AI Scientist Collapse

llm
agents
agents-in-practice

Agents in Practice #4 on coding-agent latent space, a new benchmark for test-writing agents, and a personal anecdote about AI Scientist collapse.

Jul 9, 2026
Ryan Lee
Preview graphic for Agents in Practice #3

Agents in Practice #3: hacking benchmarks, failure attribution, and token-profiling trajectories

llm
agents
agents-in-practice

Agents in Practice #3 on hacking well-known agent benchmarks without solving tasks, a new benchmark for failure attribution, and a personal anecdote on profiling trajectories for token usage.

Jul 2, 2026
Ryan Lee
Preview graphic for Agents in Practice #2

Agents in Practice #2: Specialized Repo Explorer, Agentic Resource Discovery, and Personal Benchmark

llm
agents
agents-in-practice

Agents in Practice #2 on the efficiency of specialized repo explorers for coding agents, the Agentic Resource Discovery protocol for discovering agentic resources, and a personal anecdote about benchmarking agentic models.

Jun 25, 2026
Ryan Lee
Preview graphic for Agents in Practice #1

Agents in Practice #1: Learning from Rollouts, Long-Horizon Coding, and Memory Guardrails

llm
agents
agents-in-practice

Agents in Practice #1 on agents learning from rollouts, long-horizon coding benchmarks, and an anecdote about keeping agent memory from becoming always-loaded clutter.

Jun 18, 2026
Ryan Lee
Preview graphic for githubtocolab.

githubtocolab: Open GitHub Notebooks in Colab

tools
colab
github

A URL shortcut, converter, and bookmarklet for opening GitHub-hosted Jupyter notebooks in Google Colab.

May 26, 2026
Ryan Lee
Preview graphic for Hermes Agent as My Day-to-Day AI Interface.

Hermes Agent as My Day-to-Day AI Interface

llm
agents
hermes-agent

How Ryan Lee uses Hermes Agent as a day-to-day AI interface for context, tools, GitHub work, and personal automation.

May 21, 2026
Ryan Lee
Preview graphic for LLM Wiki with Opinions.

LLM Wiki with Opinions: Subjective Context as a First-Class Source

llm
agents
llm-wiki

A practical take on maintaining an opinionated LLM wiki as a first-class source of durable personal and project context.

May 20, 2026
Ryan Lee
Preview graphic for BrowseComp.

BrowseComp: A Simple Yet Challenging Benchmark for Browsing Agents

llm
agents
retrieval
benchmark

A practical summary of BrowseComp, a browsing-agent benchmark built around difficult multi-hop public web questions.

Aug 26, 2025
Ryan Lee
Preview graphic for Mixtral of Experts.

Mixtral of Experts

llm
attention

A concise reading note on Mixtral of Experts and sparse mixture-of-experts routing for efficient large language models.

Aug 21, 2025
Ryan Lee
Preview graphic for Mistral 7B.

Mistral 7B

llm
attention

A concise reading note on Mistral 7B, including grouped-query attention, sliding-window attention, and rolling buffer cache ideas.

Aug 10, 2025
Ryan Lee
Preview graphic for NEXT-EVAL.

NEXT-EVAL: Next Evaluation of Traditional and LLM Web Data Record Extraction

extraction
web-extraction
llm
benchmarks

A concise reading note on NEXT-EVAL, a benchmark for evaluating web data record extraction systems.

Jun 20, 2025
Ryan Lee
Preview graphic for XPath Agent.

XPath Agent: Automating Web Scraping with LLM‑Built XPaths

extraction
web-extraction
llm

A concise reading note on XPath Agent, an LLM-based system for generating durable web scraping XPaths.

Jun 19, 2025
Ryan Lee

RL Weekly 42: Special Issue on NeurIPS 2020 Competitions

reinforcement-learning
rl-weekly

In this special issue, we look at the four RL competitions that is a part of NeurIPS 2020.

Jul 3, 2020
Seungjae Ryan Lee

RL Weekly 41: Adversarial Policies, Image Augmentation, and Self-Supervised Exploration with World Models

reinforcement-learning
rl-weekly

In this issue, we look at adversarial policy learning, image augmentation in RL, and self-supervised exploration through world models.

Jun 22, 2020
Seungjae Ryan Lee

How to Setup Azure Data Science Virtual Machine (DSVM)

azure
data-science
machine-learning
Azure Data Science Virtual Machine is a virtual machine image that is built for data science. It has all the popular data science tools that you might need preinstalled and…
Jun 5, 2020
Seungjae Ryan Lee

RL Weekly 40: Catastrophic Interference and Policy Evaluation Networks

reinforcement-learning
rl-weekly

In this issue, we look at two papers combating catastrophic interference. Memento combats interference by training two independent agents where the second agent takes off when the first agent is finished. D-NN and TC-NN reduce interference by mapping the input space to a higher-dimensional space. We also look at Policy Evaluation Network, a network that predicts the expected return given a policy.

Mar 30, 2020
Seungjae Ryan Lee

RL Weekly 39: Intrinsic Motivation for Cooperation and Amortized Q-Learning

reinforcement-learning
rl-weekly

In this issue, we look at using intrinsic rewards to encourage cooperation in two-agent MDP. We also look at replacing maximization in Q-learning over all actions with a neural network to use Q-learning in environments with big action spaces.

Feb 21, 2020
Seungjae Ryan Lee

Reinforcement Learning Papers Accepted to ICLR 2020

reinforcement-learning

I have compiled a list of 106 reinforcement learning papers accepted to ICLR 2020.

Dec 24, 2019
Seungjae Ryan Lee

Papers Accepted to ICLR 2020

Here is an interactive table of all ICLR 2020 papers.

Dec 24, 2019
Seungjae Ryan Lee

RL Weekly 38: Clipped objective is not why PPO works, and the Trap of Saliency maps

reinforcement-learning
rl-weekly

In this issue, we look at the effect of PPO’s code-level optimizations and the study of saliency maps in RL.

Dec 22, 2019
Seungjae Ryan Lee

RL Weekly 37: Observational Overfitting, Hindsight Credit Assignment, and Procedurally Generated Environment Suite

reinforcement-learning
rl-weekly

In this issue, we look at Google and MIT’s study on the observational overfitting phenomenon and how overparametrization helps generalization, a new family of algorithms using hindsight credit assignment by DeepMind, and a new environment suite by OpenAI consisting of procedurally generated environments.

Dec 11, 2019
Seungjae Ryan Lee

RL Weekly 36: AlphaZero with a Learned Model achieves SotA in Atari

reinforcement-learning
rl-weekly

In this issue, we look at MuZero, DeepMind’s new algorithm that learns a model and achieves AlphaZero performance in Chess, Shogi, and Go and achieves state-of-the-art performance on Atari. We also look at Safety Gym, OpenAI’s new environment suite for safe RL.

Nov 28, 2019
Seungjae Ryan Lee

RL Weekly 35: Escaping Local Optimas in Distance-based Rewards and Choosing the Best Teacher

reinforcement-learning
rl-weekly

In this issue, we look at an algorithm that use sibling trajectories to escape local optimas in distance-based shaped rewards, and an algorithm that dynamically chooses the best demonstrator teacher to train the student.

Nov 21, 2019
Seungjae Ryan Lee

RL Weekly 34: Dexterous Manipulation of the Rubik’s Cube and Human-Agent Collaboration in Overcooked

reinforcement-learning
rl-weekly

In this issue, we look at a robot hand manipulating and “solving” the Rubik’s Cube. We also look at comparative performances of human-agnostic and human-aware RL agents in Overcooked when paired with human players.

Oct 28, 2019
Seungjae Ryan Lee

RL Weekly 33: Action Grammar, the Squashing Exploration Problem, and Task-relevant GAIL

reinforcement-learning
rl-weekly

In this issue, we look at Action Grammar RL, a hierarchical RL framework that adds new macro-actions, improving performance of DDQN and SAC in Atari environments. We then look at a new algorithm that borrows just the benefits of SAC’s bounded actions to TD3 to achieve better performance. Finally, we look at an improvement to GAIL on raw pixel observations by focusing on task-relevant details.

Oct 14, 2019
Seungjae Ryan Lee

RL Weekly 32: New SotA Sample Efficiency on Atari and an Analysis of the Benefits of Hierarchical RL

reinforcement-learning
rl-weekly

In this issue, we look at LASER, DeepMind’s improvement to V-trace that achieves state-of-the-art sample efficiency in Atari environments. We also look at Google AI and UC Berkeley’s study on hierarchical RL, analyzing and isolating the reason behind the benefits of hierarchical RL.

Oct 3, 2019
Seungjae Ryan Lee

RL Weekly 31: How Agents Play Hide and Seek, Attraction-Repulsion Actor Critic, and Efficient Learning from Demonstrations

reinforcement-learning
rl-weekly

In this issue, we look at OpenAI’s work on multi-agent hide and seek and the behaviors that emerge. We also look at Mila’s population-based exploration that exceeds the performance of various TD3 and SAC baselines. Finally, we look at DeepMind’s R2D3, a new algorithm to learn from demonstrations.

Sep 25, 2019
Seungjae Ryan Lee

Reinforcement Learning Papers Accepted to NeurIPS 2019

reinforcement-learning

I have compiled a list of 184 reinforcement learning papers accepted to NeurIPS 2019.

Sep 8, 2019
Seungjae Ryan Lee

RL Weekly 30: Learning State and Action Embeddings, a New Framework for RL in Games, and an Interactive Variant of Question Answering

reinforcement-learning
rl-weekly

In this issue, we look at a representation learning method to train state and action embeddings paired with TD3. We also look at a new framework with 20+ environments and algorithms from multiple fields of RL. Finally, we look at a new take on using RL for Question Answering (QA).

Sep 2, 2019
Seungjae Ryan Lee

GSoC TensorFlow Part 7: Retrospective

reinforcement-learning
gsoc
tensorflow

As of August 27th, the Google Summer of Code coding phase is officially over. In this post, I look back at the summer, reviewing my accomplishments and shortcomings. Because I will continue contributing to TensorFlow and TF-Agents, I also outline my plans for the next fall.

Aug 28, 2019
Seungjae Ryan Lee

RL Weekly 29: The Behaviors and Superstitions of RL, and How Deep RL Compares with the Best Humans in Atari

reinforcement-learning
rl-weekly

In this issue, we look at reinforcement learning from a wider perspective. We look at new environments and experiments that are designed to test and challenge the agents’ capabilities. We also compare existing RL agents against the playthroughs of best Atari human players.

Aug 19, 2019
Seungjae Ryan Lee

GSoC TensorFlow Part 6: Evaluating RND on Mountain Car

reinforcement-learning
gsoc
tensorflow

After finishing implementing Random Network Distillation by Burda et al., now it is time to evaluate the algorithm in various hard exploration environments. I first start the evaluation on Mountain Car, a simple environment that requires extensive exploration to reach the goal state.

Aug 13, 2019
Seungjae Ryan Lee

RL Weekly 28: Free-Lunch Saliency and Hierarchical RL with Behavior Cloning

reinforcement-learning
rl-weekly

This week, we first look at Free-Lunch Saliency, a built-in interpretability module that does not deteriorate performance. Then, we look at HRL-BC, a combination of high-level RL policy with low-level skills trained through behavior cloning.

Aug 12, 2019
Seungjae Ryan Lee

RL Weekly 27: Diverse Trajectory-conditioned Self Imitation Learning and Environment Probing Interaction Policies

reinforcement-learning
rl-weekly

This week, we look at a self imitation learning method that imitates diverse past experience for better exploration. We also summarize an environment probing policy that helps an agent adapt to different environments.

Aug 5, 2019
Seungjae Ryan Lee

RL Weekly 26: Transfer RL with Credit Assignment and Convolutional Reservoir Computing for World Models

reinforcement-learning
rl-weekly

This week, we summarize a new transfer learning method using the Transformer reward model, and a world model controller that does not require training the feature extraction.

Jul 22, 2019
Seungjae Ryan Lee

Setting up code-server on GCP: VSCode on Browser for Remote Work!

Visual Studio Code (VS Code) is a great code editor, but it cannot be used remotely… or can it? Code-server is VS Code running on a remote server accessible through the browser.

Jul 17, 2019
Seungjae Ryan Lee

RL Weekly 25: Replacing Bias with Adaptive Methods, Batch Off-policy Learning, and Learning Shared Model for Multi-task RL

reinforcement-learning
rl-weekly

In this issue, we focus on replacing inductive bias with adaptive solutions (DeepMind), learning off-policy from expert experience (Google Brain), and learning a shared model for multitask RL (Stanford).

Jul 15, 2019
Seungjae Ryan Lee

GSoC TensorFlow Part 5: Implementing the Core of RND

reinforcement-learning
gsoc
tensorflow

This week, I give a brief summary of Random Network Distillation (RND), complemented with the code I have written for TF-Agents. I then list further work needed to finish implementing RND, and plans for evaluating the algorithm once finished.

Jul 12, 2019
Seungjae Ryan Lee

RL Weekly 24: Benchmarks for Model-based RL and Bonus-based Exploration Methods

reinforcement-learning
rl-weekly

This week, we summarize two benchmark papers. The first paper benchmarks 11 model-based RL algorithms in 18 continuous control environments, and the second paper benchmarks 4 bonus-based exploration methods in 9 Atari environments. Both papers agree that a standardized benchmark is needed for an objective analysis of new algorithms.

Jul 8, 2019
Seungjae Ryan Lee

RL Weekly 23: Decentralized Hierarchical RL, Deep Conservative Policy Iteration, and Optimistic PPO

reinforcement-learning
rl-weekly

This week, we first introduce a ensemble of primitives without a high-level meta-policy that can make decentralized decisions. We then look at an deep learning extension of Conservative Policy Iteration that borrows the idea of DQN. Finally, we look at Optimistic PPO, an extension of PPO that encourages exploration through uncertainty bellman equation.

Jul 1, 2019
Seungjae Ryan Lee

GSoC TensorFlow Part 4: First Evaluation

reinforcement-learning
gsoc
tensorflow

This week, I look back at the first coding phase of GSoC, summarizing my work and setting goals for the next phase.

Jun 27, 2019
Seungjae Ryan Lee

RL Weekly 22: Unsupervised Learning for Atari, Model-based Policy Optimization, and Adaptive-TD

reinforcement-learning
rl-weekly

This week, we first look at ST-DIM, an unsupervised state representation learning method from MILA and Microsoft Research. We also check UC Berkeley’s new policy optimization method that uses model-based branch rollouts. Finally, we look at Adapative-TD, a new method of mixing MC and TD from DeepMind, Google Research and Universitat Pompeu Fabr.

Jun 24, 2019
Seungjae Ryan Lee

RL Weekly 21: The interplay between Experience Replay and Model-based RL

reinforcement-learning
rl-weekly

This week, we introduce three papers on replay-based RL and model-based RL. The first paper introduces SoRB, a way to combine experience replay and planning. The second paper introduces a consistency loss to ensure that a model is consistent with the real environment. The final paper compares model-based agents with replay-based agents.

Jun 17, 2019
Seungjae Ryan Lee

RL Weekly 20: Minecraft Competition, Off-policy Policy Evaluation via Classification, and Soft-attention Agent for Interpretability

reinforcement-learning
rl-weekly

This week, we introduce MineRL, a new RL competition using human priors to solve Minecraft. We also introduce OPE, a method of off-policy evaluation through classification, and a soft-attention agent for greater interpretability.

Jun 11, 2019
Seungjae Ryan Lee

GSoC TensorFlow Part 3: Simple Environment Wrapper with gin-config

reinforcement-learning
gsoc
tensorflow

This week, I implemented a simple environment wrappers to prepare myself before implementing curiosity modules.

May 30, 2019
Seungjae Ryan Lee

RL Weekly 19: Curious Object-Based Search Agent, Multiplicative Compositional Policies, and AutoRL

reinforcement-learning
rl-weekly

This week, we introduce combining unsupervised learning, exploration, and model-based RL; learning composable motor skills; and evolving rewards.

May 27, 2019
Seungjae Ryan Lee

GSoC TensorFlow Part 2: Improving Documentation

reinforcement-learning
gsoc
tensorflow

A great way to learn the material is to make modifications. This week, I summarize my experience of creating a pull request to TF-Agents to improve its documentation.

May 17, 2019
Seungjae Ryan Lee

RL Weekly 18: Survey of Domain Randomization Techniques for Sim-to-Real Transfer, and Evaluating Deep RL with ToyBox

reinforcement-learning
rl-weekly

This week, we introduce a survey of Domain Randomization Techniques for Sim-to-Real Transfer and ToyBox, a suite of redesigned Atari Environments for experimental evaluation of deep RL.

May 13, 2019
Seungjae Ryan Lee

GSoC TensorFlow Part 1: Setting Up TF-Agents

reinforcement-learning
gsoc
tensorflow

I have been accepted to Google Summer of Code program to work on TensorFlow for three months. I will be working on TensorFlow’s reinforcement learning library TF-Agents. In this post, I briefly summmarize the steps I took to setup the TF-Agents environment for future reference.

May 9, 2019
Seungjae Ryan Lee

RL Weekly 17: Information Asymmetry in KL-regularized Objective, Real-world Challenges to RL, and Fast and Slow RL

reinforcement-learning
rl-weekly

In this issue, we summarize the use of information asymmetry in KL regularized objective to regularize the policy, the challenges of deploying deep RL into real-world systems, and possible insights into psychology and neuroscience from deep RL.

May 7, 2019
Seungjae Ryan Lee

Collapsible Code Blocks in GitHub Pages

blogging

Here is a quick guide on using collapsible code blocks in GitHub pages. This might be useful when there is a large output that might be useful only to a few people reading the post.

May 2, 2019
Seungjae Ryan Lee

Using TensorBoard with PyTorch 1.1 or 1.2

pytorch

Since PyTorch 1.1, tensorboard is now natively supported in PyTorch. This post contains detailed instuctions to install tensorboard.

May 2, 2019
Seungjae Ryan Lee

RL Weekly 16: Why Performance Plateaus May Occur, and Compressing DQNs

reinforcement-learning
rl-weekly

In this issue, we introduce ‘ray interference,’ a possible cause of performance plateaus in deep reinforcement learning conjectured by Google DeepMind. We also introduce a network distillation method proposed by researchers at Carnegie Mellon University.

Apr 30, 2019
Seungjae Ryan Lee

RL Weekly 15: Learning without Rewards: from Active Queries or Suboptimal Demonstrations

reinforcement-learning
rl-weekly

In this issue, we introduce VICE-RAQ by UC Berkeley and T-REX by UT Austin and Preferred Networks. VICE-RAQ trains a classifier to infer rewards from goal examples and active querying. T-REX learns reward functions from suboptimal demonstrations ranked by humans.

Apr 22, 2019
Seungjae Ryan Lee

RL Weekly 14: OpenAI Five and Berkeley Blue

reinforcement-learning
rl-weekly

In this week’s issue, we summarize the Dota 2 match between OpenAI Five and OG eSports and introduce Blue, a new low-cost robot developed by the Robot Learning Lab at UC Berkeley.

Apr 15, 2019
Seungjae Ryan Lee

RL Weekly 13: Learning to Toss, Learning to Paint, and How to Explain RL

reinforcement-learning
rl-weekly

In this week’s issue, we summarize results from Princeton, Google, Columbia, and MIT on training a robot arm to throw objects. We also look at a model-based DDPG developed by Peking University and Megvii that can reproduce pictures through paint strokes. Finally, we look at an empirical study by Oregon State University about explaining RL to layman.

Apr 1, 2019
Seungjae Ryan Lee

RL Weekly 12: Atari Demos with Human Gaze Labels, New SOTA in Meta-RL, and a Hierarchical Take on Intrinsic Rewards

reinforcement-learning
rl-weekly

This week, we look at a new demo dataset of Atari games that include trajectories and human gaze. We also look at PEARL, a new meta-RL method that boasts sample efficiency and performance superior to previous state-of-the-art algorithms. Finally, we look at a novel method of incorporating intrinsic rewards.

Mar 25, 2019
Seungjae Ryan Lee

RL Weekly 11: The Bitter Lesson by Richard Sutton, the Promise of Hierarchical RL, and Exploration with Human Feedback

reinforcement-learning
rl-weekly

In this issue, we first look at a diary entry by Richard S. Sutton (DeepMind, UAlberta) on Compute versus Clever. Then, we look at a post summarizing Hierarchical RL by Yannis Flet-Berliac (INRIA SequeL). Finally, we summarize a paper incorporating human feedback for exploration from Delft University of Technology.

Mar 19, 2019
Seungjae Ryan Lee

RL Weekly 10: Learning from Playing, Understanding Multi-agent Intelligence, and Navigating in Google Street View

reinforcement-learning
rl-weekly

In this issue, we look at Google Brain’s algorithm of learning by playing, DeepMind’s thoughts on multi-agent intelligence, and DeepMind’s new navigation environment using Google Street View data.

Mar 10, 2019
Seungjae Ryan Lee

RL Weekly 9: Sample-efficient Near-SOTA Model-based RL, Neural MMO, and Bottlenecks in Deep Q-Learning

reinforcement-learning
rl-weekly

In this issue, we look at SimPLe, a model-based RL algorithm that achieves near-state-of-the-art results on Arcade Learning Environments (ALE). We also look at Neural MMO, a new multiagent environment by OpenAI, and an empirical analysis of possible sources of error in deep Q-learning by BAIR.

Mar 5, 2019
Seungjae Ryan Lee

RL Weekly 8: World Discovery Models, MuJoCo Soccer Environment, and Deep Planning Network

reinforcement-learning
rl-weekly

In this issue, we introduce World Discovery Models and MuJoCo Soccer Environment from Google DeepMind, and PlaNet from Google.

Feb 26, 2019
Seungjae Ryan Lee

Obstacle Tower 6: Submitting a Random Agent

reinforcement-learning
obstacle-tower
competition

We submit a random agent to the Obstacle Tower Challenge that just began.

Feb 12, 2019
Seungjae Ryan Lee

RL Weekly 7: Obstacle Tower Challenge, Hanabi Learning Environment, and Spinning Up Workshop

reinforcement-learning
rl-weekly

This week, we introduce the Obstacle Tower Challenge, a new RL competition by Unity, Hanabi Learning Environment, a multi-agent environment by DeepMind, and Spinning Up Workshop, a workshop hosted by OpenAI.

Feb 11, 2019
Seungjae Ryan Lee

Obstacle Tower 4: Understanding the Baselines

reinforcement-learning
obstacle-tower
competition

We briefly introduce Rainbow and PPO, the two baselines that was tested on Obstacle Tower.

Feb 4, 2019
Seungjae Ryan Lee

Obstacle Tower 5: Possible Improvements to the Baselines

reinforcement-learning
obstacle-tower
competition

We play the Obstacle Tower game to understand the qualities of a successful agent.

Feb 4, 2019
Seungjae Ryan Lee

Slow Papers: The Obstacle Tower: A Generalization Challenge in Vision, Control, and Planning (Juliani et al., 2019)

slowpapers
reinforcement-learning

The rapid pace of research development in Deep Reinforcement Learning has been driven by the presence of fast and challenging simulation environments. These environments often take the form of video games, such as the Atari games provided in the Arcade Learning Environment (ALE). In the past year, however, significant progress has been made in achieving superhuman performance on even the most difficult and heavily studied game in the ALE: Montezumas Revenge. We propose a new benchmark environment, Obstacle Tower: a high visual fidelity, 3D, 3rd person, procedurally generated environment. An agent in the Obstacle Tower must learn to solve both low level control and high-level planning problems in tandem learning from pixels and a sparse reward signal in order to make it as high as possible up the tower. In this paper we outline the environment and provide a set of initial baseline results using current state of the art Deep RL methods as well as human players. In all cases these algorithms fail to produce agents capable of performing anywhere near human level on a set of evaluations designed to test both memorization and generalization ability. As such, we believe that the Obstacle Tower has the potential to serve as a helpful Deep RL benchmark now and into the future.

Feb 2, 2019
Seungjae Ryan Lee

Obstacle Tower 3: Observation Space and Action Space

reinforcement-learning
obstacle-tower
competition

We analyze the observation space and the action space provided by the Obstacle Tower environment.

Jan 31, 2019
Seungjae Ryan Lee

Obstacle Tower 2: Playing the Game

reinforcement-learning
obstacle-tower
competition

We play the Obstacle Tower game to understand the qualities of a successful agent.

Jan 30, 2019
Seungjae Ryan Lee

Obstacle Tower 1: Installing the Environment

reinforcement-learning
obstacle-tower
competition

Unity introduced the Obstacle Tower Challenge, a new reinforcement learning contest with a difficult environment. In this post, we guide the readers on installing the environment on Linux using conda.

Jan 29, 2019
Seungjae Ryan Lee

RL Weekly 6: AlphaStar, Rectified Nash Response, and Causal Reasoning with Meta RL

reinforcement-learning
rl-weekly

This week, we look at AlphaStar, a Starcraft II AI, PSRO_rN, an evaluation algorithm encouraging diverse population of well-trained agents, and a novel Meta-RL approach for causal reasoning. All three results are from DeepMind.

Jan 28, 2019
Seungjae Ryan Lee

Deep RL Seminar Week 2: Deep Q-Networks

reinforcement-learning
deep-rl-seminar

This week, we reviewed various improvements made to the Deep Q-Network algorithm.

Jan 22, 2019
Seungjae Ryan Lee

RL Weekly 5: Robust Control of Legged Robots, Compiler Phase-Ordering, and Go Explore on Sonic the Hedgehog

reinforcement-learning
rl-weekly

This week, we look at impressive robust control of legged robots by ETH Zurich and Intel, compiler phase-ordering by UC Berkeley and MIT, and a partial implementation of Uber’s Go Explore.

Jan 21, 2019
Seungjae Ryan Lee

RL Weekly 4: Generating Problems with Solutions, Optical Flow with RL, and Model-free Planning

reinforcement-learning
rl-weekly

In this issue, we introduce new curriculum learning algorithm by Uber AI Labs, model-free planning algorithm by DeepMind, and optical-flow based control algorithm by Intel Labs and University of Freiburg.

Jan 14, 2019
Seungjae Ryan Lee

RL Weekly 3: Learning to Drive through Dense Traffic, Learning to Walk, and Summarizing Progress in Sim-to-Real

reinforcement-learning
rl-weekly

In this issue, we introduce the DeepTraffic competition from Lex Fridman’s MIT Deep Learning for Self-Driving Cars course. We also review a new paper on using SAC to control a four-legged robot, and introduce a website summarizing progress in sim-to-real algorithms.

Jan 7, 2019
Seungjae Ryan Lee

PyTorch Implementations of Policy Gradient Methods

reinforcement-learning
policy-gradient
pytorch

A well-written baseline is crucial to research. We compare and recommend popular open source implementations of reinforcement learning algorithms in PyTorch.

Dec 28, 2018
Seungjae Ryan Lee

RL Weekly 2: Tuning AlphaGo, Macro-strategy for MOBA, Sim-to-Real with conditional GANs

reinforcement-learning
rl-weekly

In this issue, we discuss hyperparameter tuning for AlphaGo from DeepMind, Hierarchical RL model for a MOBA game from Tencent, and GAN-based Sim-to-Real algorithm from X, Google Brain, and DeepMind.

Dec 24, 2018
Seungjae Ryan Lee

RL Weekly 1: Soft Actor-Critic Code Release; Text-based RL Competition; Learning with Training Wheels

reinforcement-learning
rl-weekly

In this inaugural issue of the RL Weekly newsletter, we discuss Soft Actor-Critic (SAC) from BAIR, the new TextWorld competition by Microsoft Research, and AsDDPG from University of Oxford and Heriot-Watt University.

Dec 17, 2018
Seungjae Ryan Lee

Slow Papers: Exploration by Random Network Distillation (Burda et al., 2018)

reinforcement-learning
slowpapers

We introduce an exploration bonus for deep reinforcement learning methods that is easy to implement and adds minimal overhead to the computation performed. The bonus is the error of a neural network predicting features of the observations given by a fixed randomly initialized neural network. We also introduce a method to flexibly combine intrinsic and extrinsic rewards. We find that the random network distillation (RND) bonus combined with this increased flexibility enables significant progress on several hard exploration Atari games. In particular we establish state of the art performance on Montezuma’s Revenge, a game famously difficult for deep reinforcement learning methods. To the best of our knowledge, this is the first method that achieves better than average human performance on this game without using demonstrations or having access the underlying state of the game, and occasionally completes the first level. This suggests that relatively simple methods that scale well can be sufficient to tackle challenging exploration problems.

Nov 1, 2018
Seungjae Ryan Lee

Slow Papers: A Deeper Look at Experience Replay (Zhang and Sutton, 2017)

reinforcement-learning
slowpapers

Recently experience replay is widely used in various deep reinforcement learning (RL) algorithms, in this paper we rethink the utility of experience replay. It introduces a new hyper-parameter, the memory buffer size, which needs carefully tuning. However unfortunately the importance of this new hyper-parameter has been underestimated in the community for a long time. In this paper we did a systematic empirical study of experience replay under various function representations. We showcase that a large replay buffer can significantly hurt the performance. Moreover, we propose a simple O(1) method to remedy the negative influence of a large replay buffer. We showcase its utility in both simple grid world and challenging domains like Atari games.

Oct 23, 2018
Seungjae Ryan Lee

Slow Papers: Neural Fitted Q Iteration (Riedmiller, 2005)

reinforcement-learning
slowpapers

This paper introduces NFQ, an algorithm for efficient and effective training of a Q-value function represented by a multi-layer perceptron. Based on the principle of storing and reusing transition experiences, a model-free, neural network based RL algorithm is proposed. The method is evaluated on three benchmark problems. It is shown empirically, that reasonably few interactions with the plant are neeed to generate control policies of high quality.

Oct 2, 2018
Seungjae Ryan Lee

AI for Prosthetics Week 9 - 10: Unorthodox Approaches

reinforcement-learning
ai-for-prosthetics
competition

We end the series by exploring possible unorthodox approaches for the competition. These are approaches that deviate from the popular policy gradient methods such as DDPG or PPO.

Sep 12, 2018
Seungjae Ryan Lee

Notes from the ai.x 2018 Conference: Faster Reinforcement Learning via Transfer

reinforcement-learning
conference

SK T-Brain hosted the ai.x Conference on September 6th at Seoul, South Korea. At this conference, John Schulman (OpenAI) spoke about faster reinforcement learning via transfer.

Sep 10, 2018
Seungjae Ryan Lee

Pommerman 1: Understanding the Competition

reinforcement-learning
competition

Pommerman is one of NIPS 2018 Competition tracks, where the participants seek to build agents to compete against other agents in a game of Bomberman. In this post, we simply explain the basics of Pommerman, leaving reinforcement learning to later posts.

Sep 2, 2018
Seungjae Ryan Lee

AI for Prosthetics Week 6: General Techniques of RL

reinforcement-learning
ai-for-prosthetics
competition

This week, we take a step back from the competition and study common techniques used in Reinforcement Learning.

Aug 13, 2018
Seungjae Ryan Lee

AI for Prosthetics Week 5: Understanding the Reward

reinforcement-learning
ai-for-prosthetics
competition

The goal of reinforcement learning is defined by the reward signal - to maximize the cumulative reward throughout an episode. In some ways, the reward is the most important aspect of the environment for the agent: even if it does not know about values of states or actions (like Evolutionary Strategies), if it can consistently get high return (cumulative reward), it is a great agent.

Jul 25, 2018
Seungjae Ryan Lee

AI for Prosthetics Week 3-4: Understanding the Observation Space

reinforcement-learning
ai-for-prosthetics
competition

The observation can be roughly divided into five components: the body parts, the joints, the muscles, the forces, and the center of mass. For each body part component, the agent observes its position, velocity, acceleration, rotation, rotational velocity, and rotational acceleration.

Jul 18, 2018
Seungjae Ryan Lee

AI for Prosthetics Week 2: Understanding the Action Space

reinforcement-learning
ai-for-prosthetics
competition

Last week, we saw how a valid action has 19 numbers, each between 0 and 1. The 19 numbers represented the amount of force to put to each muscle. I know barely anything about muscles, so I decided to manually go through all the muscles to understand the effects of each muscle…

Jul 4, 2018
Seungjae Ryan Lee

AI for Prosthetics Week 1: Understanding the Challenge

reinforcement-learning
ai-for-prosthetics
competition

The AI for Prosthetics challenge is one of NIPS 2018 Competition tracks. In this challenge, the participants seek to build an agent that can make a 3D model of human with prosthetics run. This challenge is a continuation of the Learning to Run challenge (shown below) that was part of NIPS 2017 Competition Track. The challenge was enhanced in three ways…

Jun 27, 2018
Seungjae Ryan Lee

Jupyter Notebook extensions to enhance your efficiency

Jupyter Notebook is a great tool that allows you to integrate live code, equations, visualizations and narrative text into a document. It is used extensively in data science. However, for developers who have used IDEs with abundant features, the simplicity of Jupyter Notebook might be problematic.

Jun 19, 2018
Seungjae Ryan Lee

Bias-variance Tradeoff in Reinforcement Learning

reinforcement-learning

Bias-variance tradeoff is a familiar term to most people who learned machine learning. In the context of Machine Learning, bias and variance refers to the model: a model that underfits the data has high bias, whereas a model that overfits the data has high variance. In Reinforcement Learning, we consider another bias-variance tradeoff.

Jun 10, 2018
Seungjae Ryan Lee

I learned DQNs with OpenAI competition

reinforcement-learning
competition

On April, OpenAI held a two-month-long competition called the Retro Contest where participants had to develop an agent that can achieve perform well on unseen custom-made stages of Sonic the Hedgehog. The agents were limited to 100 million steps per stage and 12 hours of time on a VM with 6 E5-2690v3 cores, 56GB of RAM, and a single K80 GPU.

Jun 5, 2018
Seungjae Ryan Lee

Effective Data: Partition

machine-learning

To train a good model, you need lots of data. Luckily, over the last few decades, collecting data has become much easier. However, there is little value to data if you use it incorrectly. Even if you double or triple the dataset manually or through data augmentation, without proper partition of data, you will be left clueless on how helpful adding more data was.

Apr 14, 2018
Seungjae Ryan Lee
No matching items