Reading List

Ilya’s Paper List

Fundamentals

  • The Annotated Transformer

  • The First Law of Complexodynamics

  • The Unreasonable Effectiveness of RNNs

  • Understanding LSTM Networks

  • Recurrent Neural Network Regularization

  • Keeping Neural Networks Simple by Minimizing the Description Length of the Weights

  • Pointer Networks

  • ImageNet Classification with Deep CNNs

  • Order Matters: Sequence to Sequence for Sets

  • GPipe: Efficient Training of Giant Neural Networks

  • Deep Residual Learning for Image Recognition

  • Multi-Scale Context Aggregation by Dilated Convolutions

  • Neural Quantum Chemistry

  • Attention Is All You Need

  • Neural Machine Translation by Jointly Learning to Align and Translate

  • Identity Mappings in Deep Residual Networks

  • A Simple NN Module for Relational Reasoning

  • Variational Lossy Autoencoder

  • Relational RNNs

  • Quantifying the Rise and Fall of Complexity in Closed Systems

  • Neural Turing Machines

  • Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

  • Scaling Laws for Neural LMs (arxiv.org)

  • A Tutorial Introduction to the Minimum Description Length Principle (arxiv.org)

  • Machine Super Intelligence Dissertation (vetta.org)

  • PAGE 434 onwards: Komogrov Complexity (lirmm.fr)

  • CS231n Convolutional Neural Networks for Visual Recognition (cs231n.github.io)

Additional Resources

Search and Recommender Systems

Tag

Title

Survey; CTR

Deep Learning for Click-Through Rate Estimation

Survey; Embedding

Embedding in Recommender Systems: A Survey

Survey; Ranking

A Survey on Accuracy-Oriented Neural Recommendation: From Collaborative Filtering to Information-Rich Recommendation

Survey; Retrieval

A Comprehensive Survey on Retrieval Methods in Recommender Systems

Survey; Objective

Self-Supervised Learning for Recommender Systems: A Survey

Method; CTR

Factorization Machines

Method; CTR

Wide & Deep Learning for Recommender Systems

Method; CTR

DeepFM: A Factorization-Machine based Neural Network for CTR Predictio

Method; CTR

xDeepFM: Combining Explicit and Implicit Feature Interactions for Recommender Systems

Method; CTR

Deep & Cross Network for Ad Click Predictions

Method; CTR

DCN V2: Improved Deep & Cross Network and Practical Lessons for Web-scale Learning to Rank Systems

Method; CTR

Deep Interest Network for Click-Through Rate Prediction

Method; Feedback Delay

Addressing Delayed Feedback for Continuous Training with Neural Networks in CTR prediction

Must Read Papers

Tag

Title

Attention

MHA: Attention Is All You Need

Attention

MQA: Fast Transformer Decoding: One Write-Head is All You Need

Attention

GQA: Training Generalized Multi-Query Transformer Models from Multi-Head Checkpoints

Attention

Attention is Turing Complete

Decoding

Fast Inference from Transformers via Speculative Decoding

Activation

GLU Variants Improve Transformer

Norm

Layer Normalization

Norm

Root Mean Square Layer Normalization

PE

RoFormer: Enhanced Transformer with Rotary Position Embedding

MLM

BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding

MLM

RoBERTa: A Robustly Optimized BERT Pretraining Approach

MLM

TwinBERT: Distilling Knowledge to Twin-Structured BERT Models for Efficient Retrieval

MLM

Finetuned Language Models Are Zero-Shot Learners

RTD

Electra: Pre-training Text Encoders as Discriminators Rather Than Generators

CLM

GPT2: Language Models are Unsupervised Multitask Learners

CLM

GPT3: Language Models are Few-Shot Learners

CLM

LLaMA: Open and Efficient Foundation Language Models

CLM

LLaMA 2: Open Foundation and Fine-Tuned Chat Models

CLM

LLaMA 3: The Llama 3 Herd of Models

MoE

Mixtral: Mixtral of Experts

MoE

DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning

PLM

XLNet: Generalized Autoregressive Pretraining for Language Understanding

GLM

GLM: General Language Model Pretraining with Autoregressive Blank Infilling

MoE

Switch Transformers: Scaling to Trillion Parameter Models with Simple and Efficient Sparsity

MoE

OLMoE: Open Mixture-of-Experts Language Models

Seq2Seq

BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation Translation and Comprehension

Seq2Seq

T5: Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer

Multilingual

XLM: Cross-lingual Language Model Pretraining

Multilingual

XLM-R: Unsupervised Cross-lingual Representation Learning at Scale

Multilingual

mT5: A Massively Multilingual Pre-trained Text-to-Text Transformer

Generalisation

Grokking: Generalization Beyond Overfitting on Small Algorithmic Datasets

Scaling

Scaling Laws for Neural Language Models

Scaling

Scaling Laws for Autoregressive Generative Modeling

Scaling

Scaling Laws for Data Filtering – Data Curation cannot be Compute Agnostic

Contrastive

E5: Text Embeddings by Weakly-Supervised Contrastive Pre-training

Contrastive

Unsupervised Dense Information Retrieval with Contrastive Learning

IR

Dense Passage Retrieval for Open-Domain Question Answering

IE:NER

UniversalNER: Targeted Distillation from Large Language Models for Open Named Entity Recognition

LLM

Aligning language models to follow instructions

LLM

Scaling Instruction-Finetuned Language Models

LLM

InstructGpt: Training language models to follow instructions with human feedback

LLM

Injecting New Knowledge into Large Language Models via Supervised Fine-Tuning

LLM

The Instruction Hierarchy: Training LLMs to Prioritize Privileged Instructions

LLM

Self-instruct: Aligning language model with self generated instructions

LLM

PPO: Proximal Policy Optimization Algorithms

LLM

SFT+RLHF: Learning to summarize from human feedback

LLM

Reflexion: Language Agents with Verbal Reinforcement Learning

LLM

RLCD: Reinforcement Learning from Contrastive Distillation for Language Model Alignment

LLM

On-Policy Distillation of Language Models: Learning from Self-Generated Mistakes

LLM

DPO: Direct Preference Optimization: Your Language Model is Secretly a Reward Model

LLM

Understanding Reference Policies in Direct Preference Optimization

LLM

D2PO: Discriminator-Guided DPO with Response Evaluation Models

LLM

Aligning with Human Judgement: The Role of Pairwise Preference in Large Language Model Evaluators

LLM

RAG: Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks

LLM

Large Language Models Are Latent Variable Models

LLM

DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models

Quant

LLM.int8(): 8-bit Matrix Multiplication for Transformers at Scale

Quant

LoRA Low-Rank Adaptation of Large Language Models

Quant

QLORA: Efficient Finetuning of Quantized LLMs

Quant

SmoothQuant: Accurate and Efficient Post-Training Quantization for Large Language Models

DiT

Scalable Diffusion Models with Transformers

DiT

Scaling Rectified Flow Transformers for High-Resolution Image Synthesis

ViT

Patch n’ Pack: NaViT - a Vision Transformer for any Aspect Ratio and Resolution

ViT

Long Context Transfer from Language to Vision

Multimodal

Transfusion: Predict the Next Token and Diffuse Images with One Multi-Modal Model

Eval

HELM: Holistic Evaluation of Language Models

Eval

MMLU: Measuring Massive Multitask Language Understanding

Eval

MMMU: A Massive Multi-discipline Multimodal Understanding and Reasoning Benchmark for Expert AGI

Hallucination

SelfCheckGPT: Zero-Resource Black-Box Hallucination Detection for Generative Large Language Models

Hallucination

G-Eval: NLG Evaluation using GPT-4 with Better Human Alignment

Hallucination

To Believe or Not to Believe Your LLM

Representation

Scaling and evaluating sparse autoencoders

Representation

Probabilistic Topic Modelling with Transformer Representations

Representation

Matryoshka Representation Learning

Representation

Not All Language Model Features Are Linear

Context: Full

FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness

Context: Full

FlashAttention-2: Faster Attention with Better Parallelism and Work Partitioning

Context: Full

FlashAttention-3: Fast and Accurate Attention with Asynchrony and Low-precision

Context: Full

Effective Long-Context Scaling of Foundation Models

Context: Sparse

Longformer: The Long-Document Transformer

Context: Sparse

Transformer-XL: Attentive Language Models Beyond a Fixed-Length Context

Context

Ring Attention with Blockwise Transformers for Near-Infinite Context

Context

Lost in the Middle: How Language Models Use Long Contexts

Long Context

ChatQA 2: Bridging the Gap to Proprietary LLMs in Long Context and RAG Capabilities

Long Context

LongLoRA: Efficient Fine-tuning of Long-Context Large Language Models

Long Context

YaRN: Efficient Context Window Extension of Large Language Models

Long Context

Data Engineering for Scaling Language Models to 128K Context

Long Context

Leave No Context Behind: Efficient Infinite Context Transformers with Infini-attention

Long Context

LongAlign: A Recipe for Long Context Alignment of Large Language Models

Long Context

Chain of Agents: Large Language Models Collaborating on Long-Context Tasks

Memory

MemoryBank: Enhancing Large Language Models with Long-Term Memory

Memory

Augmenting Language Models with Long-Term Memory

Memory

Recurrent Memory Transformer

Memory

Scaling Transformer to 1M tokens and beyond with RMT

Memory

Beyond Attention: Breaking the Limits of Transformer Context Length with Recurrent Memory

KG

Language Models as Knowledge Bases?

KG

Language Models are Open Knowledge Graphs

KG

Unifying Large Language Models and Knowledge Graphs: A Roadmap

KG

QA-GNN: Reasoning with Language Models and Knowledge Graphs for Question Answering

KG

SimKGC: Simple Contrastive Knowledge Graph Completion with Pre-trained Language Models

KG

Give Us the Facts: Enhancing Large Language Models with Knowledge Graphs for Fact-aware Language Modeling

KG

Knowledge Graph Large Language Model (KG-LLM) for Link Prediction

ML Design General Principle

  • [Andrew Ng] Machine Learning Yearning

  • [Chip Huyen] Designing Machine Learning Systems

  • [Burkov] Machine Learning Engineering

  • [Jeff Smith] Machine Learning Systems

  • [Lakshmanan] Machine Learning Design Patterns

  • [UCB] System Design for Large Scale Machine Learning

ML Math

ML Algorithms

  • [Naumann] The Art of Differentiating Computer Programs

Applied ML

MLE Papers

  • [ACM] DNN for YouTube Recommendations

  • [FB] Local Search

  • [FB] Photo Search

  • [FB] Recommeding items to more than a billion people

  • [ICML] ScaNN

  • [NeurIPS] DiskANN

  • [KDD] Predicting Clicks on Ads at Facebook

  • [RecSys] Recommending What Video to Watch Next

  • 91% of ML Models Degrade in Time

ML Interviews

  • [Kashan] Deep Learning Interviews

System Design General Principles

System Design Algorithms