:: CS Student Archive ::

Content Index

• Attention Is All You Need (Paper) - Introduces the Transformer, replacing recurrence with self-attention to enable parallel training and superior sequence modeling.
• BERT: Pre-training of Deep Bidirectional Transformers (Paper) - Masked language modeling and next-sentence prediction pre-train deep bidirectional Transformers, dramatically improving NLP benchmarks.
• Language Models are Few-Shot Learners (GPT-3) (Paper) - Shows large autoregressive models can perform diverse tasks via in-context learning without gradient updates.
• Deep Residual Learning for Image Recognition (ResNet) (Paper) - Residual connections enable very deep convolutional networks by mitigating vanishing gradients, achieving state-of-the-art on ImageNet.
• Generative Adversarial Networks (GANs) (Paper) - Introduces adversarial training: a generator and discriminator play a minimax game to produce realistic samples.
• ImageNet Classification with Deep CNNs (AlexNet) (Paper) - Demonstrates deep convolutional networks with ReLU, dropout, and data augmentation, igniting the deep learning revolution in vision.
• You Only Look Once (YOLO) (Paper) - Single-stage object detection that frames detection as regression, enabling real-time performance.
• Mastering the game of Go with deep neural networks and tree search (AlphaGo) (Paper) - Combines policy/value networks with Monte Carlo tree search to surpass human performance in Go.
• Learning Transferable Visual Models From Natural Language Supervision (CLIP) (Paper) - Pre-trains vision encoders on image–text pairs to enable zero-shot recognition and flexible multimodal tasks.
• NeRF: Representing Scenes as Neural Radiance Fields (Paper) - Optimizes a radiance field with volumetric rendering to synthesize novel views from sparse images.
• Chain-of-Thought Prompting Elicits Reasoning in LLMs (Paper) - Simple prompting that encourages step-by-step reasoning improves performance on arithmetic and logic tasks.
• FlashAttention: Fast and Memory-Efficient Exact Attention (Paper) - IO-aware attention algorithm that tiles queries/keys to maximize GPU bandwidth and reduce memory reads.