I am a Ph.D. candidate in Computer Science at University of Wisconsin–Madison, advised by Fred Sala in the Sprocket Lab, where I work on the science of foundation model scaling, data-efficiency, and adaptation to high-impact scientific domains—all with the ultimate goal of developing powerful scientific research agents.

I have completed research internships at Meta’s Llama team (working on scaling laws with Dieuwke Hupkes), Together AI (hybrid language models with Tri Dao), and Microsoft Research (Physics of AGI group with Sébastien Bubeck). I received an honorable mention for the Jane Street Graduate Research Fellowship (2025) and was named an MLCommons Rising Star (2023).

My academic path began at Fresno City College before earning my B.S. at UC San Diego where I worked with Gary Cottrell and Sanjoy Dasgupta and M.S. at Carnegie Mellon University with Ameet Talwalkar and Zack Lipton.

My research develops the principles of a general “science of foundation models,” with the overarching goal of developing powerful scientific research agents. I focus on three interconnected areas:

1. understanding the end-to-end foundation model design pipeline from scientific principles,
2. developing data efficient foundation model training and adaptation techniques, and
3. adapting foundation model pipelines to non-standard scientific domains.

Through scaling laws research with Meta, I have shown that knowledge acquisition and reasoning tasks require fundamentally different compute allocation strategies—challenging the assumption of uniform scaling. I complement this with automated ML techniques that adapt general models to domains like protein folding and climate modeling, where labeled data is scarce and expert knowledge is critical. My goal is enabling scientists to leverage state-of-the-art AI without requiring extensive ML expertise, accelerating discovery across fields from biology to physics.

Email: nick11roberts [at] cs [dot] wisc [dot] edu
Office: Morgridge Hall 5548

• Excited to visit Chris Ré’s lab at Stanford University this August!
• Two papers accepted to COLM 2025!
  • Manticore for constructing pretrained hybrids from non-hybrid base models without pretraining!
  • Meta MLGym: a framework for benchmarking and developing agents for AI research!
• Presented at ACL 2025:
  • We discover a new scaling law phenomenon! Compute optima for knowledge favor larger models, whereas those for reasoning favor more data.
• I am extremely fortunate to have received an honorable mention for the Jane Street Graduate Research Fellowship! Thank you, Jane Street!

Fresh off the Press

• Breakeven complexity: A new perspective on evaluating neural partial differential equation solvers
  Yijing Zhang, Nicholas Roberts, Tanya Marwah, Mikhail Khodak.
  Preprint.
  

• Pareto Optimal Code Generation
  Gabriel Orlanski, Nicholas Roberts, Aws Albarghouthi, Frederic Sala.
  Preprint.
  [arXiv]

• Procedural Generation of Algorithm Discovery Tasks in Machine Learning
  Alexander Goldie, Zilin Wang, Adrian Hayler, Deepak Nathani, Edan Toledo, Aleksandra Kalisz, Ken Thampiratwong, Michael Beukman, Alistair Letcher, Shashank Reddy Chirra, Clarisse Wibault, Theo Wolf, Charles O'Neill, Uljad Berdica, Nicholas Roberts, Saeed Rahmani, Roberta Raileanu, Shimon Whiteson, Jakob Foerster.
  Preprint.
  

Conference Publications

• Pretrained Hybrids with MAD Skills
  Nicholas Roberts, Samuel Guo, Zhiqi Gao, Satya Sai Srinath Namburi GNVV, Sonia Cromp, Chengjun Wu, Chengyu Duan, Frederic Sala.
  COLM 2025.
  [Paper] [arXiv]

• MLGym: A New Framework and Benchmark for Advancing AI Research Agents
  Deepak Nathani, Lovish Madaan, Nicholas Roberts, Nikolay Bashlykov, Ajay Menon, Vincent Moens, Amar Budhiraja, Despoina Magka, Vladislav Vorotilov, Gaurav Chaurasia, Dieuwke Hupkes, Ricardo Silveira Cabral, Tatiana Shavrina, Jakob Foerster, Yoram Bachrach, William Yang Wang, Roberta Raileanu.
  COLM 2025.
  [Paper] [arXiv]

• Compute Optimal Scaling of Skills: Knowledge vs Reasoning
  Nicholas Roberts, Niladri Chatterji, Sharan Narang, Mike Lewis, Dieuwke Hupkes.
  ACL Findings 2025.
  [Paper] [arXiv]

• Stronger Than You Think: Benchmarking Weak Supervision on Realistic Tasks
  Tianyi Zhang*, Linrong Cai*, Jeffrey Li, Nicholas Roberts, Neel Guha, Frederic Sala.
  NeurIPS 2024.
  [Paper] [arXiv]

• Geometry-Aware Adaptation for Pretrained Models
  Nicholas Roberts, Xintong Li, Dyah Adila, Sonia Cromp, Tzu-Heng Huang, Jitian Zhao, Frederic Sala.
  NeurIPS 2023.
  [Paper] [arXiv]

• Skill-it! A data-driven skills framework for understanding and training language models
  Mayee Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré.
  NeurIPS 2023 (spotlight).
  [Paper] [arXiv]

• Generative Modeling Helps Weak Supervision (and Vice Versa)
  Benedikt Boecking, Nicholas Roberts, Willie Neiswanger, Stefano Ermon, Frederic Sala, Artur Dubrawski.
  ICLR 2023.
  [Paper] [arXiv] [Code]

• AutoWS-Bench-101: Benchmarking Automated Weak Supervision with 100 Labels
  Nicholas Roberts*, Xintong Li*, Tzu-Heng Huang, Dyah Adila, Spencer Schoenberg, Cheng-Yu Liu, Lauren Pick, Haotian Ma, Aws Albarghouthi, Frederic Sala.
  NeurIPS 2022.
  [Paper] [arXiv] [Code] [Blog]

• NAS-Bench-360: Benchmarking Neural Architecture Search on Diverse Tasks
  Renbo Tu*, Nicholas Roberts*, Mikhail Khodak, Junhong Shen, Frederic Sala, Ameet Talwalkar.
  NeurIPS 2022.
  [Paper] [arXiv] [Code] [Website] [Blog]

• Lifting Weak Supervision To Structured Prediction
  Harit Vishwakarma, Nicholas Roberts, Frederic Sala.
  NeurIPS 2022.
  [Paper] [arXiv]

• Universalizing Weak Supervision
  Changho Shin, Winfred Li, Harit Vishwakarma, Nicholas Roberts, Frederic Sala.
  ICLR 2022.
  [Paper] [arXiv]

• Rethinking Neural Operations for Diverse Tasks
  Nicholas Roberts*, Mikhail Khodak*, Tri Dao, Liam Li, Christopher Ré, Ameet Talwalkar.
  NeurIPS 2021.
  [Paper] [arXiv] [Code] [Software Package] [Talk]
  
  Preliminary version: Searching for Convolutions and a More Ambitious NAS
  Nicholas Roberts*, Mikhail Khodak*, Tri Dao, Liam Li, Maria-Florina Balcan, Christopher Ré, Ameet Talwalkar.
  AAAI 2021 Workshop on Learning Network Architecture During Training (plenary talk).
  

• Learning from Discriminative Feature Feedback
  Sanjoy Dasgupta, Akansha Dey, Nicholas Roberts, Sivan Sabato.
  NeurIPS 2018.
  [Paper]

Journal Publications

• Tabby: Tabular Adaptation for Language Models
  Sonia Cromp, Satya Sai Srinath Namburi GNVV, Catherine Cao, Mohammed Alkhudhayri, Samuel Guo, Nicholas Roberts, Frederic Sala
  Transactions on Machine Learning Research (TMLR) 2025.
  [Paper]

• Beyond the Imitation Game: Quantifying and extrapolating the capabilities of language models
  Aarohi Srivastava, ..., Nicholas Roberts (276), ..., (442 authors).
  Transactions on Machine Learning Research (TMLR) 2023 (Finalist for Outstanding Certification).
  ICLR 2025.
  [arXiv] [Code]
  

• NL-Augmenter: A Framework for Task-Sensitive Natural Language Augmentation
  Kaustubh D. Dhole, ..., Nicholas Roberts (85), ..., (128 authors).
  Northern European Journal of Language Technology (NEJLT) 2023.
  [arXiv] [Code]
  

• AutoML Decathlon: Diverse Tasks, Modern Methods, and Efficiency at Scale
  Nicholas Roberts*, Samuel Guo*, Cong Xu*, Ameet Talwalkar, David Lander, Lvfang Tao, Linhang Cai, Shuaicheng Niu, Jianyu Heng, Hongyang Qin, Minwen Deng, Johannes Hog, Alexander Pfefferle, Sushil Ammanaghatta Shivakumar, Arjun Krishnakumar, Yubo Wang, Rhea Sukthanker, Frank Hutter, Euxhen Hasanaj, Tien-Dung Le, Mikhail Khodak, Yuriy Nevmyvaka, Kashif Rasul, Frederic Sala, Anderson Schneider, Junhong Shen, Evan Sparks
  PMLR NeurIPS 2022 Competition Track.
  [Paper] [Website] [Submission Site] [Code] [Blog]

• Small Molecule Accurate Recognition Technology (SMART) to Enhance Natural Products Research
  Chen Zhang*, Yerlan Idelbayev*, Nicholas Roberts, Yiwen Tao, Yashwanth Nannapaneni, Brendan M. Duggan, Jie Min, Eugene C. Lin, Erik C. Gerwick, Garrison W. Cottrell, William H. Gerwick.
  Scientific Reports 2017.
  [Paper]
  
  Poster: Small Molecule Accurate Recognition Technology (SMART): A Digital Frontier to Reshape Natural Product Research
  Chen Zhang*, Yerlan Idelbayev*, Nicholas Roberts (presenter), Yiwen Tao, Yashwanth Nannapaneni, Brendan M. Duggan, Jie Min, Eugene C. Lin, Erik C. Gerwick, Garrison W. Cottrell, William H. Gerwick.
  Best Spotlight Presentation Award: Applied Machine Learning Days 2018.
  

Workshop Publications and Preprints

• R&B: Domain Regrouping and Data Mixture Balancing for Efficient Foundation Model Training
  Albert Ge, Tzu-Heng Huang, John Cooper, Avi Trost, Ziyi Chu, Satya Sai Srinath Namburi GNVV, Ziyang Cai, Kendall Park, Nicholas Roberts, Frederic Sala.
  ICML 2025 DIG-BUGS Workshop.
  ICML 2025 DataWorld Workshop.
  [arXiv]

• MoRe Fine-Tuning with 10x Fewer Parameters
  Wenxuan Tan, Nicholas Roberts, Tzu-Heng Huang, Jitian Zhao, John Cooper, Samuel Guo, Chengyu Duan, Frederic Sala.
  ICML 2024 Efficient Systems for Foundation Models (ES-FoMo) Workshop.
  ICML 2024 Workshop on Foundation Models in the Wild.
  [arXiv]

• Understanding Neural Architecture Search by its Architecture Parameters
  Nicholas Roberts, Yingyu Liang, Frederic Sala.
  Midwest Machine Learning Symposium 2023.
  

• ScriptoriumWS: A Code Generation Assistant for Weak Supervision
  Tzu-Heng Huang, Harit Vishwakarma, Catherine Cao, Spencer Schoenberg, Nicholas Roberts, Frederic Sala.
  ICLR 2023 Deep Learning for Code Workshop.
  

• AutoML for Climate Change: A Call to Action
  Renbo Tu, Nicholas Roberts, Vishak Prasad, Sibasis Nayak, Paarth Jain, Frederic Sala, Ganesh Ramakrishnan, Ameet Talwalkar, Willie Neiswanger, Colin White.
  NeurIPS 2022 Tackling Climate Change with Machine Learning Workshop.
  [arXiv]

• Decoding and Diversity in Machine Translation
  Nicholas Roberts, Davis Liang, Graham Neubig, Zachary C. Lipton.
  NeurIPS 2020 Resistance AI Workshop.
  [arXiv]

• A Simple Setting for Understanding Neural Architecture Search with Weight-Sharing
  Mikhail Khodak, Liam Li, Nicholas Roberts, Maria-Florina Balcan, Ameet Talwalkar.
  ICML 2020 AutoML Workshop.
  [Paper]

• Weight-Sharing Beyond Neural Architecture Search: Efficient Feature Map Selection and Federated Hyperparameter Tuning
  Mikhail Khodak*, Liam Li*, Nicholas Roberts, Maria-Florina Balcan, Ameet Talwalkar.
  MLSys 2020 On-Device Intelligence Workshop.
  [Paper]

• Deep Connectomics Networks: Neural Network Architectures Inspired by Neuronal Networks
  Nicholas Roberts, Dian Ang Yap, Vinay U. Prabhu.
  NeurIPS 2019 Real Neurons and Hidden Units Workshop.
  [arXiv]

• Using Deep Siamese Neural Networks to Speed up Natural Products Research
  Nicholas Roberts, Poornav S. Purushothama, Vishal T. Vasudevan, Siddarth Ravichandran, Chen Zhang, William H. Gerwick, Garrison W. Cottrell.
  NeurIPS 2019 workshop on Machine Learning and the Physical Sciences.
  [Paper]

• Grassmannian Packings in Neural Networks: Learning with Maximal Subspace Packings for Diversity and Anti-Sparsity
  Dian Ang Yap, Nicholas Roberts, Vinay U. Prabhu.
  NeurIPS 2019 workshop on Bayesian Deep Learning.
  NeurIPS 2019 workshop on Information Theory and Machine Learning.
  [arXiv]

• Model Weight Theft With Just Noise Inputs: The Curious Case of the Petulant Attacker
  Nicholas Roberts, Vinay U. Prabhu, Matthew McAteer.
  ICML 2019 workshop on Security and Privacy of Machine Learning (spotlight).
  [arXiv]

University of Wisconsin–Madison

Ph.D. Computer Science
Mathematics minor

August 2021 - Present

• Visiting student with Chris Ré's lab at Stanford University (August 2025)
  
• Member of the Wisconsin Triathlon Team
  
• Member of the Hoofer Sailing Club
  
• Scratch Club volunteer (teaching CS to Madison area 4th-5th graders)
  

Carnegie Mellon University

M.S. Machine Learning

August 2019 - May 2021

• MSML Student Committee Leader
  
• MSML Admissions Committee Member
  

University of California, San Diego

B.S. Computer Science
Mathematics minor
CSE Honors Program

September 2015 - March 2019
Magna Cum Laude and CSE Highest Distinction honors

• Tutor for Data Science 10 (Principles of Data Science)
  
• Tutor for Data Science 20 (Data Structures for Data Science)
  
• Data analyst, Triton Engineering Student Council
  
• Machine learning workshop facilitator, Data Science Student Society at UCSD
  
• House leader, Tau Beta Pi, California Psi
  

Fresno City College

Computer Science
Leon S. Peters Honors Program

August 2013 - May 2015

• Tutor for CIT 65 (Android Application Development)
  
• Mathematics Tutor
  
• Computer Science Tutor
  
• President/Founder, Google Developer Group Fresno City College
  
• Treasurer, Science and Engineering Club
  

Meta AI

Research Scientist Intern 🦙
(London, UK)

• Llama Generative AI pretraining team with Dieuwke Hupkes
• Discovered a new phenomenon: knowledge and reasoning skills have different scaling behavior
• Incorporated the AUP score, originally used in our AutoML Decathlon competition, into Meta MLGym
• Technologies used: Python, PyTorch

Together AI

Research Intern 🐍
(San Francisco, CA)

• Research with Tri Dao
• Analyzed the role of specific attention heads for long-range retrieval tasks in hybrid LLMs
• Investigated the mechanisms behind what makes hybrid LLMs good at in-context recall
• Technologies used: Python, PyTorch, HuggingFace

Microsoft Research

Research Intern 🦄
(Redmond, WA)

• Physics of AGI research group led by Sébastien Bubeck
• Developed activation function search techniques for large-scale LLM pretraining
• Developed learning curve extrapolation techniques to ablate architectural choices in transformers
• Technologies used: Python, PyTorch, HuggingFace

Amazon AI

Applied Scientist Intern
(Seattle, WA)

• AWS Transcribe research group led by Katrin Kirchhoff
• Researched and developed methods for hypothesis rescoring in ASR systems using neural language modeling
• Identified areas for improvement in many existing ASR systems when recognizing rare or zero shot entities
• Technologies used: Python, PyTorch, RWTH ASR, Kaldi, AWS

UnifyID

AI Fellow
Machine Learner Intern
(Redwood City, CA)

• UnifyID research lab led by Vinay Uday Prabhu
• Researched various ways in which research from network neuroscience could be applied to deep learning
• Developed a novel model extraction attack against deep learning models for computer vision using just noise inputs
• Technologies used: Python, Keras, PyTorch, TensorFlow, MATLAB, AWS

Intuit

Software Engineering Intern
(Mountain View, CA)

• Intuit Technology Futures research group
• Researched and implemented a novel deep learning model for controllable text generation as a service within Intuit
• Developed a system for proposing alternative candidate sentences for Intuit content writers using deep learning
• Investigated the use of dynamic topic models for customer support tickets to gain actionable insights over time
• Technologies used: Python, PyTorch, TensorFlow, Gensim, Keras

Altum

Applied Scientist Intern
(La Jolla, CA)

• Developed language model to extract NLP features from text data regarding cryptocurrency trading
• Investigated unsupervised learning techniques for extracting sentiment data in real time from online forums
• Technologies used: Python, PyTorch

Teradata

Software Engineering Intern
(San Diego, CA)

• Developed open source Spark-Teradata connector forked from Databricks’ connector for AWS Redshift in Scala
• Designed and implemented Teradata stored procedures in Java to mimic Redshift’s UNLOAD and COPY using S3
• Improved training methodology and architecture of deep learning time series model used internally
• Implemented system for updating the time series dataset and fine tuning the deep learning model
• Technologies used: Scala, Java, Maven, Teradata SQL, AWS, Tensorflow, Flask

The Comeback Community

Volunteer Full Stack Developer
(Remote)

• Developed site in Go, gohtml, and CSS on Google App Engine
• Mentored new developers in web development
• Technologies used: Go, Google App Engine, gohtml, HTML5, CSS3, JavaScript

Skqrl

Software Engineering Intern
(La Jolla, CA)

• Developed web crawler to compile needfinding and product data using Scrapy and Selenium
• Designed and implemented an extensible product search solution designed to handle future user search needs
• Technologies used: Python, Scrapy, Selenium, Django, MySQL, JavaScript

ModSpot

Software Engineering Intern
(Remote)

• Implemented new user account, edit profile, and login designs in Objective-C for iOS application
• Refactored analytics code for gathering statistics on app usage, helping designers make more informed choices
• Technologies used: Objective-C, Cocoa Touch, Flurry Analytics

Extracurricular interests:

• Aspiring triathlete
• Sailing
• Pottery
• Guitar
• Interior design
• A budding interest in plants
• Longboard construction and woodworking

Extra-Extracurricular interests:

• Into vintage lever espresso machines, and proud owner of a 1974 Olympia Cremina (sans asbestos, of course).
• Ordained Dudeist priest. I can legally officiate weddings in the US. If you’d like to book me for your wedding, please feel free to reach out.
• Head Researcher of Margarita Machine Lounge Therapy at Vacation Inc. Check out our selection of luxury sunscreens today! And use my referral link!¹
• Unofficial Toyota Prius landspeed record holder at Bonneville Speedway.
• Trying to get into Iceboat racing. Personal goal: become the fastest Iceboat racer to ever hail from Fresno, CA (a low bar, indeed).
• Did I mention I’m into espresso? I added a microcontroller to a cheap espresso machine to elevate it to the level of a $4,000 Decent.

Credentials:

Photograph of me shredding, circa 2009:

Corn, circa long ago, colorized:

Mixtape:

¹Don’t forget to check out Poolsuite FM: the ultra-summer music player for the Macintosh Computer; transporting you to a virtual vacation where the sun never sets.