Parsimonious Scientific machine learning

Accelerating Scientific Discovery with AI Under Data and Compute Constraints

(ParSciML-Symposium-AAAI-FSS26)

November 5 to 7, 2026

Westin, Arlington | Virginia, USA

Room: TBD

Call For Participation

Overview

Scientific discovery is increasingly constrained not by the lack of simulation tools, but by the immense computational, data, and energy costs required to explore complex scientific systems at scale. Modern AI methods have demonstrated extraordinary capabilities in large-data regimes, yet many frontier scientific domains—such as climate modeling, fluid dynamics, materials discovery, fusion, subsurface systems, and biological modeling—operate in precisely the opposite setting: sparse observations, expensive simulations, limited experimental access, and strict reliability requirements.

Goal

The primary goal of the symposium is to unite researchers working on efficient, physically grounded, and resource-aware AI methods for scientific discovery. The symposium will highlight emerging advances in areas such as data-efficient learning, hybrid neuro-mechanistic modeling, scientific foundation models, uncertainty-aware AI, operator learning, and compute-aware reasoning for scientific applications. Ultimately, the symposium seeks to catalyze a new research community around parsimonious scientific machine learning (Parsimonious SciML) and establish a roadmap for scalable scientific AI that prioritizes efficiency, trustworthiness, and scientific utility rather than brute-force computation alone. This bridge builds upon the success of three previous symposia organized on a related topic of Knowledge-Guided Machine Learning at the AAAI bridge program in 2024, 2025 and 2026. Accepted papers will appear in AAAI proceedings.

Topics

We encourage participation on a range of topics exploring the synergy between machine learning and scientific applications, including (but not limited to):

AI/ML techniques that generalize under training data paucity.
(Equation Discovery) Methods that uncover simple, mechanistic representations (e.g., in the form of equations) of complex unknown dynamics purely from data.
(Inverse Modeling) Efficient AI-native solutions to accelerate the outer loop problem (e.g., inverse design, inverse estimation of PDE parameters).
(Active Learning, Continual Learning) Methods that incrementally learn from sequentially generated scientific data.
(Sample-Efficient adaptation of Foundation Models) Generalizing large pre-trained models to a broad range of related down-stream scientific tasks, each with a low volume of training data.
Novel methods that:
- minimize rollout error and spectral bias in SciML surrogates.
- overcome the curse of simulation dimensionality of 2D and 3D simulations (e.g., multi-fidelity or hybrid modeling).
- improve extrapolation ability of scientific ML models, to unseen domains.
- overcome catastrophic failures of physics-informed machine learning models in challenging scientific domains.
- improve sample-efficiency and computational efficiency of scientific machine learning techniques.
Use of Scientific Knowledge:
- to improve sample efficiency and generalization under data paucity.
- as loss functions or hard constraints in the training of ML models for supervised, unsupervised, and semi-supervised applications.
- to design deep learning architectures to generate explainable and physically meaningful feature representations.
- in the design, pretraining, or finetuning of Foundation models in science.
(Data Assimilation) Use of simulated data generated by science-based models along with observations in ML frameworks.
LLMs for accelerating scientific discovery.

Submission Requirements

We are accepting two types of submissions, namely short submissions (maximum 2 pages excluding references) as extended abstracts, proposals, full paper submissions (maximum 6 pages excluding references), in a variety of tracks such as:

Lecture-style Tutorials Track (LS): We welcome short proposals on lecture-style tutorials that survey or provide new perspectives of a research area in Parsimonious SciML, with a discussion of prior literature in the area and opportunities for future research. Proposals should include an overview of the research topics that will be covered, target audience and required background knowledge, intended length of the tutorial (ranging from 20 minutes to 1 hour) and tentative format, supporting references, and prior qualifications of the team in the context of delivering related tutorials.
Hands-on Tutorials Track (HT): The goal of this track is to impart practical understanding of state-of-the-art research methodologies in Parsimonious SciML by working through examples and demonstrating the application in real-world use cases. This includes demonstrations of Parsimonious SciML algorithms, benchmark datasets, tools, code-bases, or coding platforms in an interactive hands-on format that is easy to follow for a broad audience. Proposals submitted to this track should include an overview of the tutorial topics, target audience, required background knowledge and software requirements, intended length of the tutorial (ranging from 45 minutes to 1.5 hours) and tentative format, supporting references, and prior qualifications of the team in the context of delivering related tutorials.
Early Career Lightning Talks Track (EC): We want to promote next-generation leaders in the field of scientific machine learning (SciML) including graduate students, postdocs, and early career investigators by giving them an opportunity to present 5-minute lightning talks on their research at our event. Submissions should include a description of the research goals and prior work of the researcher in SciML, their motivation for attending the symposium, and a short author bio.
Papers Track:
- Regular Papers (RP): Early stage or detailed experimental investigations of topics related to scientific machine learning.
- Blue Sky Idea Papers (BS): New perspectives or unexplored ideas in scientific machine learning.
- Datasets and Benchmarks Papers (DB): Novel scientific datasets, evaluations, and benchmarks.
- Dissertation Forum Papers (DF): Forum to present graduate student dissertation research on scientific machine learning.

The title of the submission should clearly state which one of the seven tracks is being targeted by adding a prefix including the two-letter track code -- shown in parenthesis above -- followed by a colon before the title of the submission. For example, if the paper title of a paper submitted to the regular paper track is "Sample-Efficient Neural Surrogates for PDE Modeling" , the submission title should be "RP: Sample-Efficient Neural Surrogates for PDE Modeling".

Submissions should be formatted according to the AAAI-27 template (two-column, camera-ready style; see Author Kit) and submitted via EasyChair. Accepted papers will appear in AAAI proceedings. Posters will also be showcased for accepted papers. Please feel free to reach out to the organizing committee if you have any questions about the submission instructions.

Link to EasyChair Submission Site

Key Dates

Paper Submission Deadline: September 10th, 2026, 11:59 PM (Anywhere on Earth)
Acceptance/Rejection Decision: September 20th, 2026, 11:59 PM (Anywhere on Earth)
Early Registration Deadline: October 2nd, 2026