Skill UI

torchforge is Meta's advanced PyTorch-native library designed for agentic Reinforcement Learning (RL). It fundamentally separates core RL algorithms from complex distributed infrastructure concerns. This allows researchers to rapidly prototype and experiment with novel RL methods (like GRPO) using clean, PyTorch-native abstractions, while automatically handling large-scale tasks like distributed training across thousands of GPUs using Monarch and TorchTitan.

TransformerLens is the standard library for performing mechanistic interpretability research on large language models. It provides clean interfaces to inspect and manipulate transformer internals—such as attention patterns, residual streams, and MLP outputs—via activation caching and HookPoints. This tool is essential for reverse-engineering model algorithms, performing causal tracing, and studying the internal circuits responsible for model behavior.

PufferLib is a high-performance library designed for advanced reinforcement learning simulations and training. It excels in parallel environment simulation, offering optimized vectorization and native support for multi-agent systems. Utilizing algorithms like PuffeRL (optimized PPO+LSTM), it achieves millions of steps per second, significantly accelerating research and deployment across single-agent and multi-agent environments.

Stable Baselines3 is a robust, PyTorch-based library providing production-ready implementations of core reinforcement learning algorithms (PPO, SAC, DQN, etc.). It facilitates the training of single-agent RL models, supports the creation of custom Gymnasium environments, and handles advanced parallel training via vectorized environments. It is ideal for rapid prototyping and rigorous RL experiments.

End-to-end analysis toolkit for Neuropixels extracellular recordings using SpikeInterface. Supports loading SpikeGLX/Open Ephys/NWB data, preprocessing, drift correction, spike sorting with Kilosort4 and other algorithms, quality metrics computation, and AI-assisted unit curation for publication-ready neural data analysis.

Production-ready reinforcement learning library with PPO, SAC, DQN, TD3, DDPG, A2C algorithms. Use for training RL agents, creating custom environments, implementing callbacks, and optimizing training workflows. Best for single-agent RL with Gymnasium environments. Requires Python 3.10+ and PyTorch >= 2.3.

PyTorch Geometric is a standard library for Graph Neural Networks built on PyTorch. It supports node, link, and graph classification, message passing algorithms like GCN and GAT, heterogeneous graphs, and neighbor sampling. Ideal for building scalable GNN models with custom datasets and transforms.

A comprehensive guide for authors aiming to submit machine learning manuscripts to the Asian Conference on Machine Learning (ACML). This skill helps evaluate manuscript fit, re-frame research from journal or arXiv styles, understand required contribution types (e.g., algorithms, theorems, benchmarks), and align the paper's evidence bar with the expectations of a top-tier, regional AI/ML venue. Use it for venue selection and submission strategy development.

This skill provides a comprehensive guide for authors submitting papers in biomedical imaging algorithms, microscopy, and medical imaging to top-tier conferences like ISBI. It helps authors strategically frame their work, assess scope fit, ensure strong evidence, and understand the unique requirements of the venue, ensuring the manuscript is optimized for conference review standards.

This guide provides detailed best practices for ensuring the reproducibility of advanced AI research across various domains, including algorithms, theories, multi-agent systems, and datasets. It outlines the specific evidence required—from pseudocode and seeds to full proofs and public datasets—to elevate a submission's credibility rating from 'irreproducible' to 'convincing' for top-tier AI conferences.

This guide helps researchers determine the optimal venue for their Computer Vision paper, specifically addressing the unique identity of WACV (Applications-first) versus major conferences like CVPR, ICCV, and ECCV. It guides authors on whether their contribution is driven by real-world applications/constraints or by advancing core algorithms, and how to choose between the Applications or Algorithms track within WACV.

A comprehensive knowledge base covering the theoretical fundamentals and advanced algorithms of Deep Reinforcement Learning (Deep RL). It details core concepts like Bellman equations, policy gradients, and the trade-offs between policy optimization and Q-learning. The guide analyzes leading algorithms such as PPO, SAC, TRPO, and DDPG, and provides rigorous best practices for debugging and running scientifically sound RL experiments.