Towards Foundation Models for Science
Francois Lanusse, Mariel Pettee, Bruno Regaldo-Saint Blancard
on behalf of Shirley Ho and the Polymathic AI team
What are foundation models?
The Key Ideas
- Large models pre-trained on task-agnostic objectives on massive & diverse datasets
- Through fine-tuning, they can be adapted downstream for a variety of tasks/datasets
- They may contain advantageous inductive biases about particular domains or data structures
- They are actively changing workflows:
- "around 80% of the U.S. workforce could have at least 10% of their work tasks affected by the introduction of LLMs, while approximately 19% of workers may see at least 50% of their tasks impacted." (arXiv:2303.10130 [econ.GN])
Can we translate these innovations into a paradigm shift in machine learning for scientific applications?
Polymathic
Advancing Science through Multi‑Disciplinary AI
Our mission: to usher in a new class of machine learning for scientific data, building models that can leverage shared concepts across disciplines."
Meet the Polymathic AI Team
Colm-Cille
Caulfield University of Cambridge
|
Leslie
Greengard Flatiron Institute
New York University |
David
Ha Sakana AI
|
Yann
LeCun Meta AI
New York University |
---|---|---|---|
Stephane
Mallat École Normale Supérieure
Collège de France Flatiron Institute |
David
Spergel Simons Foundation |
Olga Troyanskaya Flatiron Institute Princeton University |
Laure
Zanna New York University
|
Our Resources
SCIENTIFIC ADVISORY GROUP
COMPUTING RESOURCES
- Internal H100 GPUs resources at the Flatiron Institute
- External 500k GPU hours (V100 and A100)
- In the process of securing additional O(10^2) dedicated H100 GPUs
The Foundation Model Spectrum
Language-like/less structured
Structured-data
Scientific Reasoning
Multi-Modality
Generalization to Data-Limited Domains
How can we build foundation models that jump across scientific disciplines?
- Should we treat scientific data as if we treat language?
- Should we treat scientific data the way we have been treating them in ML? (structured as in grids, images, videos, graphs, etc...)
- What is a common basis across multiple disciplines and modalities?
The Foundation Model Spectrum
Language-like/less structured
Structured-data
xVal
A Continuous Number Encoding for LLMs
AstroCLIP
Cross-Modal Pretraining for Astronomical data
MPP
Multiple Physics Pretraining for Physical Surrogate Models
Scientific Reasoning
Multi-Modality
Generalization to Data-Limited Domains
xVal
A Continuous Number Encoding for LLMs
Project led by Siavash Golkar, Mariel Pettee, Michael Eickenberg, Alberto Bietti
Accepted contribution at the NeurIPS 2023 AI4Science Workshop
See Mariel's talk from Monday
The problem: existing LLMs are not suitable for reliable zero-shot numerical operations.
arXiv:2305.18654 [cs.CL]
arXiv:2109.03137 [cs.CL]
They make erratic, discontinuous predictions.
Even fine-tuning exhaustively does not grant out-of-distribution generalization abilities.
xVal in a Nutshell: a continuous numerical encoding for language models
xVal in a Nutshell: a continuous numerical encoding for language models
This encoding strategy has 3 main benefits:
-
Continuity
-
It embeds key information about how numbers continuously relate to one another, making its predictions more appropriate for many scientific applications.
-
-
Interpolation
-
It makes better out-of-distribution predictions than other numerical encodings.
-
-
Efficiency
-
By using just a single token to represent any number, it requires less memory, compute resources, and training time to achieve strong results.
-
xVal in a Nutshell: a continuous numerical encoding for language models
xVal shows improved predictions for out-of-distribution values.
xVal in a Nutshell: a continuous numerical encoding for language models
When evaluated on multi-digit multiplication tasks, xVal performs comparably well, and is less prone to large outliers:
And when evaluated on compound operations of basic arithmetic, xVal shows the strongest performance:
xVal in a Nutshell: a continuous numerical encoding for language models
Future directions: improving the dynamic range of the embedded values.
MPP
Multiple Physics Pretraining for Physical Surrogate Models
Project led by Michael McCabe, Bruno Régaldo, Liam Parker, Ruben Ohana, Miles Cranmer
Oral presentation at the NeurIPS 2023 AI4Science Workshop
Context
- Previous works on large domain-specific pretrained models: chemistry, medicine, astrophysics, climate, ...
→ extension on surrogate modeling of spatiotemporal physical systems - Spatiotemporal predictions motivated by faster surrogates for PDE solvers, systems that are hard to simulate with current models/hardware
- Situations where data is expensive…
Time
Ex: N-body simulation
Springel et al. 2005
Can we build a foundation model that
would be finetunable on a few training examples?
Background
- Main pretraining strategies
- autoregressive prediction
- masked reconstruction
- contrastive learning - No conditioning on physical parameters
- Spatiotemporal physics: PDEs from a physical system with typically conservations and symmetries…
→ Suggests that there are some learnable shared features
Natural choice for physical surrogate modeling
Physical Systems from PDEBench
Navier-Stokes
Incompressible
Compressible
Shallow Water
Diffusion-Reaction
Takamoto et al. 2022
Compositionality and Pretraining
Architecture for MPP
Balancing objectives during training
Normalized MSE:
Experiment 1: Performance on Pretraining Tasks
Context size: 16 frames
Experiment 2: Transfer
Compressible Navier-Stokes
M = 0.1
M = 1.0
Fun Fact: Finetuning with VideoMAE works quite well…
Tube masking
ViT
ViT
ViT
Trained on reconstructing masked pixels on natural videos (SSV2 and K400)
Tong et al. 2022
Experiment 3: Broader Downstream Tasks
Regression Problems on Incompressible Navier-Stokes
Mixed results
Long time predictions on Navier-Stokes?
Conclusion
- MPP-pretraining approach with shared embedding space and autoregressive predictions
- A single transformer model matches/outperforms baselines for each task without finetuning
- Transfer capabilities were demonstrated
- Evaluation on broader downstream tasks
- Open code and pretrained models: https://github.com/PolymathicAI/multiple_physics_pretraining
AstroCLIP
Cross-Modal Pre-Training for Astronomical Foundation Models
Project led by Francois Lanusse, Liam Parker, Siavash Golkar, Miles Cranmer
Accepted contribution at the NeurIPS 2023 AI4Science Workshop
The Data Diversity Challenge
- Success of recent foundation models is driven by large corpora of uniform data (e.g LAION 5B).
- Scientific data comes with many additional challenges:
- Metadata matters
- Wide variety of measurements/observations
Credit: Melchior et al. 2021
Credit:DESI collaboration/DESI Legacy Imaging Surveys/LBNL/DOE & KPNO/CTIO/NOIRLab/NSF/AURA/unWISE
Towards Large Multi-Modal Observational Models
Most General
Most Specific
Independent models for every type of observation
Single model capable of processing all types of observations
Towards Large Multi-Modal Observational Models
Most General
Most Specific
Independent models for every type of observation
Single model capable of processing all types of observations
Bytes Are All You Need (Horton et al. 2023)
Towards Large Multi-Modal Observational Models
Most General
Most Specific
Independent models for every type of observation
Single model capable of processing all types of observations
Bytes Are All You Need (Horton et al. 2023)
AstroCLIP
What is CLIP?
Contrastive Language Image Pretraining (CLIP)
(Radford et al. 2021)
One model, many downstream applications!
Flamingo: a Visual Language Model for Few-Shot Learning (Alayrac et al. 2022)
Hierarchical Text-Conditional Image Generation with CLIP Latents (Ramesh et al. 2022)
Contrastive Learning in Astrophysics
Self-Supervised similarity search for large scientific datasets (Stein et al. 2021)
Example of Science Application: Identifying Galaxy Tidal Features
The AstroCLIP approach
- We use spectra and multi-band images as our two different views for the same underlying object.
- DESI Legacy Surveys (g,r,z) images, and DESI EDR galaxy spectra.
- Once trained, we can do example retrieval by nearest neighbor search.
How we do it
We take a two steps approach:
- Build self-supervised model separately for images and spectra
- Images: Start from pre-trained ResNet 50 from Stein et al. (2021)
- Spectra: Pretrain by mask modeling a GPT-2 like transformer on spectra
- Train an embedding module on top of each backbone under InfoNCE loss
- Images: Simple MLP
- Spectra: Cross-attention module
More examples of retrieval
Image Similarity
Spectral Similarity
Image-Spectral Similarity
Visualizing properties of the embedding space
UMAP representation of spectra embeddings
Testing Structure of Embedding Space by k-NN Regression
Comparison to image-only SSL (from Stein et al. 2021)
The Information Point of View
- The InfoNCE loss is a lower bound on the Mutual Information between modalities
Shared physical information about galaxies between images and spectra
Thinking about data from a hierarchical Bayesian model point of view
=> We are building summary statistics for the physical parameters describing an object in a completely data driven way
What comes next!
- Idea of finding a common shared representation diverse observational modalities
- Next steps would be, embedded data from different instruments, different filters, etc... to build a universal embedding for types of galaxy observations
- Next steps would be, embedded data from different instruments, different filters, etc... to build a universal embedding for types of galaxy observations
- These pretrained models will act as strong off-the-shelf model for many downstream tasks:
- Never need to train your own CNN anymore
Teaser for what comes next
We are just getting started!
Thank you for listening!
Towards Foundation Models for Science
By eiffl
Towards Foundation Models for Science
Overview talk of the Polymathic AI Initiative
- 428