Staying on Top of the Literature

"Four to six weeks in the lab can save you an hour in the library."
— G.C. Quarderer

The Challenge

The number of publications grows exponentially, making it impossible to read everything. However, staying current with relevant research remains crucial for effective scientific work.

Discovery Methods

• Google Scholar alerts on key topics, authors, and "related papers"
• Journal browsing of major venues (browse the table of content for Nature, Science, PNAS)
• Social media (especially Twitter/X for ML) as a quality filter
• Journal clubs and literature seminars
• Paper recommendation systems I use Scholar Inbox, which sends me a daily email with papers that are similar to the ones I read recently

Recommended Workflow

1. Collection System

• Use Zotero as your central repository
• Create an "Inbox" folder for quick captures
• Install browser extensions for seamless paper collection
• • Consider tools like PaperMemory for finding forgotten papers

2. Processing Routine

• Read papers directly in Zotero (desktop or mobile)
• Take highlights in the Zotero app
• I use my note-taking app (Logseq) in a second window to take notes

3. Team Sharing

• Always share interesting papers with the team using the Zulip channel
• Include a brief summary explaining relevance or key insights

Core Responsibilities

Your responsibility: Stay current with literature in your research area. While others may provide initial pointers, maintaining awareness is ultimately your job.

Best practice: Read broadly, including papers that simply spark your curiosity—unexpected connections often drive innovation.

Essential Reading List

From our group

• Our review on "general purpose models"
• Our review on evaluating machine learning systems

Core ML/AI Foundations

• Attention is All You Need - The transformer architecture (start with the Annotated Transformer)
• Neural Message Passing for Quantum Chemistry - Graph neural networks for molecules
• Automatic Chemical Design Using a Data-Driven Continuous Representation of Molecules - VAEs for chemical design
• Meta-Learning in Neural Networks: A Survey - Comprehensive meta-learning overview
• Interpretable Machine Learning - Essential reference on interpretability

Chemical/Materials Informatics

• BigSMILES: A Structurally-Based Line Notation for Describing Macromolecules - Polymer representation standard
• Polymer Genome: A Data-Powered Polymer Informatics Platform - Major polymer informatics effort
• AI-driven design of catalysts and materials for ring opening polymerization - Domain-specific language approach

Transformer Deep Dive

• The Illustrated Transformer - Visual explanation of transformers
• Leveraging LLMs for Predictive Chemistry - LLMs applied to chemistry
• LIFT: Language-Interfaced Fine-Tuning - Fine-tuning for non-language tasks
• Scaling Laws for Autoregressive Generative Modeling - Understanding model scaling
• The Power of Scale for Parameter-Efficient Prompt Tuning - Efficient fine-tuning methods

Graph Neural Networks

• JT-VAE - Junction tree variational autoencoders for molecules
• Jumping Knowledge Networks - Advanced GNN architectures

ML Theory & Ethics

• Understanding Deep Learning Requires Rethinking Generalization - Fundamental insights on generalization
• The Lottery Ticket Hypothesis - Sparse neural network training
• On the Dangers of Stochastic Parrots - Critical perspective on large language models
• AI Ethics Course - Comprehensive ethics framework

Software Development

• The Pragmatic Programmer - Essential software development principles

Basic Research Tools

Essential Skills:

• The Turing Way - Comprehensive guide to reproducible research

• The Missing Semester - Practical computing skills every researcher needs

• What Every CS Major Should Know - Foundational computer science concepts

Books

General Reading

• Greatness Cannot Be Planned - On open-endedness and serendipity in research
• Chop Wood Carry Water - Philosophy and mindset for sustained work
• Systems Science - Understanding complex systems

Software Development

• The Pragmatic Programmer - Essential development principles

Media & Ongoing Learning

Podcasts

• Dwarkesh Podcast - In-depth interviews with researchers and thought leaders

YouTube Channels

• Yannic Kilcher - ML paper reviews and "ML News" updates