Topics
RAG
Retrieval quality, graph structure, and long-context limits.
LightRAG: Simple and Fast Retrieval-Augmented Generation
GraphRAG cost, dual-level retrieval, and incremental updates
Lost in the Middle
Why models miss information placed in the middle of long context
A Graph RAG Approach to Query-Focused Summarization
Graph-based summarization over global document structure
Engineering RAG Systems for Real-World Applications
Evaluation and operational concerns for production RAG
LLM
Transformer structure, long-context behavior, and inference stability.
Attention Is All You Need
The Transformer core and how modern LLM stacks evolved from it
Context Rot: How Increasing Input Tokens Impacts LLM Performance
How longer input can reduce reliability and reasoning quality
Defeating Nondeterminism in LLM Inference
Reproducibility and nondeterminism in LLM inference
Small Language Models are the Future of Agentic AI
Why smaller models can matter in agent workflows
Generative Models
Image generation, diffusion, and visual understanding papers.
Diffusion Models Beat GANs on Image Synthesis
Why diffusion models became strong image synthesis baselines
MIRAGE: The Illusion of Visual Understanding
Limits and evaluation questions for vision-language models
Foundations
Classic deep learning papers connected back to modern model design.
ImageNet Classification with Deep Convolutional Neural Networks
AlexNet and the start of large-scale image classification
Batch Normalization
Training stability and normalization in deep networks
Deep Residual Learning for Image Recognition
How residual connections changed deep model training
Recommender Systems
Recommendation models through retrieval, ranking, and personalization.
Deep Neural Networks for YouTube Recommendations
Candidate generation and ranking in large-scale recommendation
Deep Learning based Recommender System: A Survey
Major architectures and research directions in deep recommender systems
Dynamic Multi-Behavior Sequence Modeling for Next Item Recommendation
Using multi-behavior sequences for next item recommendation