Discover the Best AI Resources

Re-examines residual blocks and shows that pure identity skip connections plus pre-activation (BN-ReLU before each conv) let gradients flow cleanly enough to train a 1001-layer ResNet, hitting 4.62% error on CIFAR-10.

PythonRobotics

Readable, minimal-dependency Python implementations of core robotics algorithms — localization (EKF, particle filter), SLAM (ICP, FastSLAM), path planning (A*, RRT*, PRM), and path tracking (LQR, MPC) — written to be studied, not just run.

Insomnia

Cross-platform API client for debugging, designing, testing and mocking GraphQL, REST, WebSockets, SSE and gRPC. Provides selectable storage backends (Local Vault, Git Sync, Cloud Sync with optional E2EE), a native OpenAPI editor, built-in test suites and a plugin ecosystem — useful for reproducible API development and pre-production validation.

SQuAD

Provides about 100,000 crowd‑written question–answer pairs from Wikipedia where each answer is a text span in the passage, used to train and evaluate extractive question‑answering models. Includes train/validation splits, span offsets, Parquet format, CC BY‑SA 4.0.

LightGBM

Trains gradient-boosted decision trees for classification, ranking, and large-scale tabular ML with lower memory use and faster training. GOSS and EFB help it handle high-dimensional sparse data on CPU, GPU, and distributed setups.

PyTorch

Lets researchers and engineers build neural networks as regular Python programs, with GPU-backed tensors, autograd, distributed training, and production paths through TorchScript and related tooling.

Crawlee

Unified Node.js library for web crawling and browser automation that fetches pages and files via headless browsers or raw HTTP. Provides persistent queues, proxy rotation, session management, storage, and human-like fingerprints to build scalable data pipelines (e.g., RAG/LLM datasets).

TensorRT

Converts trained PyTorch, TensorFlow, and ONNX models into GPU-tuned inference engines via layer fusion, kernel auto-tuning, and reduced precision. Cuts latency, raises throughput on NVIDIA GPUs from Turing (INT8), with FP8 on Ada+ and FP4 on Blackwell+.

CuPy

Provides a NumPy/SciPy-compatible GPU array library for Python, enabling existing NumPy/SciPy numerical code to run on NVIDIA CUDA and AMD ROCm with minimal changes. Exposes low-level CUDA features (RawKernels, Streams) and offers prebuilt binaries for multiple CUDA/ROCm versions.

Reframes the VAE's tendency to ignore its latent code as a controllable design choice: by limiting a PixelCNN decoder's receptive field and using autoregressive flow priors, the code is forced to keep only global structure and discard local texture.

Builds deep learning from the ground up, first teaching the linear algebra, probability, and numerical methods most ML texts assume you know. Three parts run from math foundations to practical networks to research topics, favoring reasoning over recipes.

openpilot

Upgrades supported cars with an open robotics operating system for driver assistance, combining vehicle integration, safety testing, data collection, and community development around comma.ai hardware.