Explore machine learning papers and reviews related to Deep Learning. Find insights, analysis, and implementation details.
Lianghui Zhu, Bencheng Liao, +4
Follow Vision Mamba from image patches through selective state updates, two aligned scan directions, a constant-resolution backbone, and runnable Python.
Yue Liu, Yunjie Tian, +7
Follow VMamba from a compact running state through four image scan orders, aligned cross-merge, the final VSS block, and runnable Python.
Hanyu Zhao, Zhenhua Han, +9
HiveD reserves groups of GPUs that sit together, not GPU counts, so a tenant never queues longer in a shared cluster than in a private one of the same size.
Mohammad Shoeybi, Mostofa Patwary, +4
Megatron-LM splits each transformer layer across GPUs with two all-reduces forward and two backward, and trains an 8.3B-parameter GPT-2 on 512 V100s.
Yunge Li, Lanyu Xu
Reordering image tokens along a Hilbert curve makes 2D local attention block-sparse, for about 4x faster window attention and 18x faster slide attention.
Leyla Naz Candogan, Arshia Afzal, +2
VIOLIN multiplies ViT attention by decay masks from eight space-filling curves: 1,296 extra parameters on DeiT-B and up to 8.7 points on VTAB-1K spatial tasks.
Kan Zhu, Yufei Gao, +14
How NanoFlow raises LLM serving throughput by overlapping compute, memory and network work inside a single GPU: it shows serving is compute-bound, splits each batch into nano-batches, and shares SMs between concurrent kernels, reaching 1.91x the throughput of TensorRT-LLM.
Tri Dao, Daniel Y. Fu, +3
How FlashAttention computes exact attention with linear memory by tiling Q, K, and V into SRAM-resident blocks and fusing the softmax, avoiding the quadratic HBM cost of materializing the full attention matrix.
Edward J. Hu, Yelong Shen, +6
How LoRA adapts a frozen large language model by learning a low-rank update ΔW = (α/r)·BA to each weight matrix, training under 1% of the parameters of full fine-tuning and adding zero inference latency once merged.
Woosuk Kwon, Zhuohan Li, +7
How PagedAttention (the memory manager behind vLLM) applies OS-style virtual-memory paging to the KV cache — fixed-size blocks, a block table, and copy-on-write prefix sharing — to reduce final-block waste, remove external block fragmentation, and improve throughput in the paper’s measured serving workloads.
William Fedus, Barret Zoph, +1
How the Switch Transformer scales to trillions of parameters by routing each token to a single expert (top-1 gating), decoupling model capacity from per-token compute while managing expert load with a capacity factor.
Alex Krizhevsky, Ilya Sutskever, +1
How AlexNet won ImageNet 2012 by a landslide and launched the deep-learning era: a deep convolutional network trained on two GPUs with ReLU activations, dropout regularization, and aggressive data augmentation.
Chao Jia, Yinfei Yang, +8
How ALIGN scales vision-language pre-training to 1.8 billion noisy image-alt-text pairs, showing that the sheer scale of a raw, uncurated dataset can outweigh careful data cleaning for a simple dual-encoder contrastive model.
Jacob Devlin, Ming-Wei Chang, +2
How BERT pre-trains a deep bidirectional Transformer with masked language modeling and next-sentence prediction, then fine-tunes the same model to state-of-the-art results across a wide range of NLP tasks.
Junnan Li, Dongxu Li, +2
How BLIP unifies vision-language understanding and generation in one model, and bootstraps noisy web data with CapFilt — a captioner that synthesizes captions and a filter that removes noisy image-text pairs.
Jiahui Yu, Zirui Wang, +4
How CoCa trains a single image-text foundation model with both a contrastive loss and a generative captioning loss by splitting the text decoder into unimodal and multimodal halves, computed in one forward pass.
Jean-Baptiste Alayrac, Jeff Donahue, +25
How Flamingo bridges a frozen vision encoder and a frozen language model with a Perceiver Resampler and gated cross-attention layers, enabling few-shot learning from interleaved image-text sequences.
Bin Xiao, Haiping Wu, +7
How Florence-2 turns detection, segmentation, grounding, captioning, and OCR into one sequence-to-sequence problem driven by task prompts, expressing spatial outputs as location tokens in the text vocabulary.
Ian J. Goodfellow, Jean Pouget-Abadie, +6
How GANs frame generative modeling as a two-player minimax game: a generator turns noise into samples while a discriminator learns to tell real from fake, driving the generator toward the true data distribution.
Peng Wang, Shuai Bai, +17
How Qwen2-VL perceives images and video at any resolution with naive dynamic resolution (variable visual tokens) and M-RoPE, a multimodal rotary position embedding that decomposes position into temporal, height, and width components.
Nikhila Ravi, Valentin Gabeur, +16
How SAM 2 extends promptable segmentation from images to video with a streaming memory bank and memory attention, propagating masks across frames and recovering objects after occlusion.
Xiaohua Zhai, Basil Mustafa, +2
How SigLIP replaces CLIP’s softmax contrastive loss with a pairwise sigmoid loss, removing the global softmax denominator and enabling chunked all-pairs computation and strong language-image pre-training at smaller batch sizes.
Diederik P. Kingma, Max Welling
How the variational autoencoder turns an autoencoder into a generative model: an encoder predicts a Gaussian latent distribution, the reparameterization trick makes sampling differentiable, and the ELBO balances reconstruction against a KL prior.
Jonathan Ho, Ajay Jain, +1
How diffusion models learn to generate images by reversing a gradual noising process — the foundation of Stable Diffusion, DALL-E, and modern image generation.
Haotian Liu, Chunyuan Li, +2
LLaVA paper: align LLMs with visual information through instruction tuning on image-text pairs, enabling multimodal understanding and reasoning.
Yanghao Li, Hanzi Mao, +2
Investigating the effectiveness of plain Vision Transformers as backbones for object detection and proposing modifications to improve their performance.
Joseph Redmon, Santosh Divvala, +2
Introducing YOLO, a unified, real-time object detection system that frames object detection as a single regression problem.
Mingxing Tan, Quoc V. Le
Connect EfficientNet’s compound scaling equations to its MBConv architecture, annotated PyTorch implementation, and real pretrained B0 versus B4 predictions.
Shaoqing Ren, Kaiming He, +2
Faster R-CNN explained: how Region Proposal Networks (RPN) enable near real-time object detection with shared convolutional features.
Alexander Kirillov, Eric Mintun, +3
SAM is a promptable segmentation model that can segment any object in an image using points, boxes, or text prompts with zero-shot generalization.
Junnan Li, Dongxu Li, +2
BLIP-2 leverages frozen image encoders and LLMs for efficient vision-language pre-training, achieving state-of-the-art multimodal performance.
Nicolas Carion, Francisco Massa, +4
Introducing DETR, a novel end-to-end object detection framework that leverages Transformers to directly predict a set of object bounding boxes.
Alexey Dosovitskiy, Lucas Beyer, +10
Vision Transformer explained with original paper equations, tensor shapes, runnable PyTorch code, and real pretrained image classification.
Ze Liu, Yutong Lin, +6
Swin Transformer: hierarchical Vision Transformer using shifted windows for efficient image classification, object detection, and segmentation.
Alec Radford, Jong Wook Kim, +10
CLIP explained: contrastive learning on 400M image-text pairs enables zero-shot image classification and powerful vision-language understanding.
Horace He
Deep learning performance optimization from first principles. Learn to identify compute-bound, memory-bound, and overhead bottlenecks with fusion techniques.
Ashish Vaswani, Noam Shazeer, +6
Deep dive into the Transformer architecture that revolutionized NLP. Understand self-attention, multi-head attention, and positional encoding.
Andrei Ivanov, Nikoli Dryden, +3
Analysis of transformer performance bottlenecks caused by data movement. Learn optimization strategies for memory-bound operations on GPUs.
Kaiming He, Xiangyu Zhang, +2
ResNet analysis: how skip connections and residual learning solved the degradation problem, enabling training of 100+ layer neural networks.