2026 Conference on Physics and AI: Bhargav Siddani

2026 Conference on Physics and AI: Bhargav Siddani

by Bhargav Siddani

Formal & Physical Sciences Physics PHPhysicsPHSStatistical physics
🎙 Bhargav Siddani 👥 34K 📅 June 30, 2026 ⏱ 23 min 👁 95 📄 original study 🧭 2026-08-03
Available in: English (current) Français

Keywords

flow matchingnon-Markovianstochastic particle systemscoarse-grainingDean-Kawasaki equation

Summary

Bhargav Siddani presents a method to capture non-Markovian dynamics in non-equilibrium stochastic systems using flow matching. The talk begins by motivating the need for coarse-grained stochastic partial differential equations (SPDEs) to simulate large particle systems efficiently. However, traditional coarse-grained SPDEs, such as the Dean-Kawasaki equation, assume noise uncorrelated in space and time, leading to non-Gaussian and non-Markovian effects at low particle counts and short timescales. To address this, the speaker proposes learning the flux distribution between computational cells using a flow matching generative model. The model incorporates history information (previous time steps) to capture non-Markovian effects and enforces statistical reflection symmetry to preserve system-level symmetries. The neural network uses a deep operator for initial conditions and a transformer for memory. Training is performed on equilibrium systems with varying particle numbers (1-50) and history lengths (0-10). The method is tested on a non-equilibrium system with a double-well potential, initialized with all particles in one well. Results show that the non-Markovian ML model closely matches ground truth random walker simulations, especially for higher-order moments (variance, skewness, kurtosis), while the standard SPDE produces negative particle counts and fails to capture these statistics. The speaker concludes that the approach is promising for interacting systems, though currently computationally more expensive for simple systems.

208 words

Critical Evaluation

The talk presents a novel application of flow matching to learn flux distributions in coarse-grained stochastic particle systems, addressing known limitations of traditional SPDE approaches. The motivation is clear: standard coarse-grained SPDEs assume uncorrelated noise, which breaks down at low particle counts and short timescales, leading to non-Gaussian and non-Markovian effects. The speaker identifies these issues and proposes a data-driven solution using generative models, specifically flow matching. The methodology is well-structured: they define integrated current, incorporate history via a transformer, and enforce reflection symmetry in the velocity field. The numerical experiment with a double-well potential demonstrates the model’s ability to capture rare events and higher-order moments, which are crucial for non-equilibrium dynamics. The comparison with ground truth random walker simulations and the standard Dean-Kawasaki equation is convincing, showing clear improvements in accuracy, particularly for variance, skewness, and kurtosis. The speaker also acknowledges limitations, such as computational cost for simple systems, and suggests future work on interacting systems. The presentation is technically rigorous, with clear explanations of the stochastic particle system, coarse-graining, and the flow matching framework. However, the talk is concise and lacks detailed derivations or extensive validation across diverse scenarios. The reliance on a single numerical experiment may limit generalizability. Additionally, the speaker does not discuss potential pitfalls or failure modes of the model. Overall, the work is promising and contributes to the intersection of physics and AI, but further validation and scalability studies are needed. The title accurately reflects the content, and the talk is well-suited for a specialized audience. The presence of a brief sponsorship mention does not affect the scientific content.

265 words

Title / Content Match

The title accurately reflects the content: a conference talk on physics and AI, specifically applying flow matching to stochastic particle systems.

Quality & Reliability

8/10

Presentation of original research with clear methodology, numerical experiments, and comparisons to ground truth. The work is preliminary but grounded in established stochastic processes and generative modeling. The speaker acknowledges limitations and future directions.

Key Moments

Markers derived by PSI from the transcript: the creator did not define chapters.

Cited Sources

Concurring Sources

Contribution & Novelties

The talk introduces a novel approach to capture non-Markovian dynamics in coarse-grained stochastic particle systems by learning flux distributions with flow matching. This addresses limitations of traditional SPDEs that assume uncorrelated noise, which fails at low particle counts and short timescales. The incorporation of history information and reflection symmetry in the generative model is a key innovation, enabling accurate prediction of higher-order moments and rare events.

Pour aller plus loin :

107 words

Radar Profile

The radar profile shows high scores across all dimensions, indicating a technically rigorous and well-presented talk with strong information content and reliability. The balance between quantity and quality of information is good, and the technical level is appropriate for a specialized audience.

Reliability 8/10