A comprehensive index of artificial intelligence and machine-learning research with AI-generated summaries, citation metrics, and direct links to papers and code.
Devvrit Khatri, Lovish Madaan, Rishabh Tiwari, et al.
This paper presents the first large-scale systematic study (400,000+ GPU-hours) defining a framework for predicting RL scaling in LLMs, and proposes a best-practice recipe, ScaleRL, enabling extrapolation from small runs.
Matthias Zwicker, Jussi Räsänen, Mario Botsch, et al.
A novel point rendering algorithm using homogeneous coordinates for perspective-correct splatting with EWA filtering and GPU implementation.
Unknown
CRYPTGPU is a system that leverages GPUs to enable fast privacy-preserving machine learning, addressing the performance bottleneck of cryptographic operations.
Chao Jin, Zili Zhang, Xuanlin Jiang, et al.
RAGCache reduces RAG latency by caching intermediate states in a knowledge tree across GPU/host memory and overlapping retrieval with generation via dynamic speculative pipelining.
Venkata Naga Sai Vishnu Rohit Pulipaka, Anish Katta, Deva Rohit Reddy Peddireddy
This paper shows that a native C++ ONNX Runtime inference engine can score RLHF rewards faster than PyTorch eager mode and FastAPI on CPU, but torch.compile is faster on GPU, and batching strategy matters more than language or runtime choice.
Felix Gremse, Marius Stärk, Josef Ehling, et al.
A GPU-accelerated software tool for interactive segmentation and rendering of multimodal biomedical volume data.
Norman P. Jouppi, Cliff Young, Nishant Patil, et al.
Evaluates a custom ASIC (TPU) for neural network inference, showing 15-30x speedup and 30-80x TOPS/Watt over contemporary CPUs and GPUs.
Alex Krizhevsky, Ilya Sutskever, Geoffrey E. Hinton
AlexNet achieved state-of-the-art image classification on ImageNet using a deep CNN with 60M parameters, ReLU, GPU training, and dropout.
Unknown
QLoRA enables fine-tuning of large language models on a single GPU by combining 4-bit NormalFloat quantization, double quantization, and paged optimizers, matching 16-bit performance.
Junsong Chen, Jincheng Yu, Yitong Li, et al.
SANA-Video 2.0 introduces hybrid linear-softmax attention and attention residuals to achieve softmax-level video quality with linear-complexity scaling, enabling 720p generation on a single GPU.
Hongzheng Chen, Jiahao Zhang, Yixiao Du, et al.
This paper investigates FPGA-based spatial acceleration for LLM inference, achieving up to 13.4x speedup over prior FPGA accelerators and 5.7x energy efficiency vs. A100 GPU.
Zhisheng Ye, Wei Gao, Qinghao Hu, et al.
A survey of deep learning workload scheduling in GPU datacenters, covering training and inference, objectives, resource utilization, and future directions.