Search Results for "2-plan"

VJEPA2 is a next-generation self-supervised learning framework for video that extends the “predict in representation space” idea from i-JEPA to the temporal domain. Instead of reconstructing pixels, it predicts the missing high-level embeddings of masked space-time regions using a context encoder and a slowly updated target encoder. This objective encourages the model to learn semantics, motion, and long-range structure without the shortcuts that pixel-level losses can invite. The...

DeepSpeed is an easy-to-use deep learning optimization software suite that enables unprecedented scale and speed for Deep Learning Training and Inference. With DeepSpeed you can: 1. Train/Inference dense or sparse models with billions or trillions of parameters 2. Achieve excellent system throughput and efficiently scale to thousands of GPUs 3. Train/Inference on resource constrained GPU systems 4. Achieve unprecedented low latency and high throughput for inference 5. Achieve extreme compression for an unparalleled inference latency and model size reduction with low costs DeepSpeed offers a confluence of system innovations, that has made large scale DL training effective, and efficient, greatly improved ease of use, and redefined the DL training landscape in terms of scale that is possible. ...

...The content is organized for fast review before an interview loop but is also deep enough for systematic study over weeks. Because it’s text-first and modular, it works equally well as a quick refresher or a backbone for a full study plan.

Implementation of Make-A-Video, new SOTA text to video generator from Meta AI, in Pytorch. They combine pseudo-3d convolutions (axial convolutions) and temporal attention and show much better temporal fusion. The pseudo-3d convolutions isn't a new concept. It has been explored before in other contexts, say for protein contact prediction as "dimensional hybrid residual networks". The gist of the paper comes down to, take a SOTA text-to-image model (here they use DALL-E2, but the same learning...

tf2_course provides the notebooks for the “Deep Learning with TensorFlow 2 and Keras” course authored by the same author, Aurélien Géron. It is structured as a teaching toolkit: you’ll find notebooks covering neural networks with Keras, lower-level TensorFlow APIs, data loading & preprocessing, convolutional and recurrent networks, and deployment/distribution of models. The material is intended for learners who already have foundational knowledge of ML and wish to deepen their understanding of deep learning frameworks and practices. ...

Deep Learning Papers Reading Roadmap is a widely known curated reading plan for deep learning that helps newcomers and practitioners navigate the vast literature in a structured and intentional way. It is built around several guiding principles: moving from outline to detail, from older foundational papers to state-of-the-art work, and from generic to more specialized areas while keeping a focus on impactful contributions.

...The materials emphasize code-first understanding: building small models, training them on accessible datasets, and analyzing their behavior. Each week culminates in a tangible outcome—such as a working classifier or sequence model—so progress is visible and motivating. The plan also introduces practical considerations like GPU usage, checkpoints, and debugging training dynamics. It aims to give you enough breadth to recognize common patterns and enough depth to implement them on your own problems.

LearningToCompare_FSL is a PyTorch implementation of the “Learning to Compare: Relation Network for Few-Shot Learning” paper, focusing on the few-shot learning experiments described in that work. The core idea implemented here is the relation network, which learns to compare pairs of feature embeddings and output relation scores that indicate whether two images belong to the same class, enabling classification from only a handful of labeled examples. The repository provides training and evaluation code for standard few-shot benchmarks such as miniImageNet and Omniglot, making it possible to reproduce the experimental results reported in the paper. It includes model definitions, data loading logic, episodic training loops, and scripts that implement the N-way K-shot evaluation protocol common in few-shot research. ...