arXiv subjects

Jingruo Peng

Publications and source records attributed to Jingruo Peng.

3 recordsLinked to original sources

arXiv2026

Acquiring Human-Like Data-Efficient Mechanics Prediction from Deep Reinforcement Learning

Humans can infer mechanical outcomes by learning from a few observations. This capacity for mechanics intuition is acquired in a data-efficient manner. Here, we propose a reinforcement learning framework to mimic this process, in which an agent encodes continuous physical observation parameters into its state and is trained via episodic switching across closely related observations. With merely two or three similar observations, the agent acquires robust mechanics intuition that generalizes over wide parameter ranges beyond the training data. Our method is demonstrated on the brachistochrone, a large-deformation elastic plate, and the quantum harmonic oscillator. We explain this generalization through a unified theoretical view: it is associated with cross-parameter Bellman consistency encouraged by episodic switching across neighboring task parameters, promoting approximate stationarity of the Bellman residual with respect to physical variations. This is consistent with a smooth policy that tracks a low-dimensional solution manifold underlying the continuum of tasks. Our work identifies episodic switching as a practically effective and theoretically motivated route to artificial mechanics intuition and suggests a computational analogy to data-efficient generalization in biological learners.

physics.comp-ph↗

arXiv2025

Variational Learning of Physical Intuition from a Few Observations: Charting Manifolds of Variational Physics

Humans often generalize physical outcomes from few observations, a desirable capacity known as physical intuition. We show that it can be computationally approached through charting the manifolds of variational physics. We conceive a variational learning framework, where small neural networks trained from merely two or three examples delineate an observation-spanned solution manifold, enabling generalization in unseen scenarios. As demonstrated across classical and quantum regimes including strongly correlated molecules, generalization emerges far beyond the training conditions. This generalization is explained by a unified theory: the Euler-Lagrange operator must be stationary with respect to the observation parameters along the solution manifold, whose complexity predicts a critical network capacity below which generalization fails. Our framework establishes a principled route to solving families of variational problems from a few instances, and shows how physical intuition can be approached through harnessing variational physics.

physics.comp-ph↗

arXiv2025

Developing Artificial Mechanics Intuitions from Extremely Small Data

Humans can possess good mechanics intuitions by learning from a few examples, which leads to the question of how to develop artificial mechanics intuitions that can be learned from small data, as we are eagerly entering the era of artificial intelligence. We propose in this Letter the sample-switchable training method, which successfully develops highly-accurate artificial mechanics intuitions that can master brachistochrone problem, catenary problem, and large nonlinear deformation problem of elastic plate by learning from no more than three samples. The model's intuitive prediction ability increases nonlinearly with respect to the number of training samples, suggesting that superb mechanics intuitions can be in-principle achieved based on a finite number of samples, reflecting how human brains form good mechanics intuitions just by learning a few cases. Our current work presents an alternative perspective for educating artificial intelligence capable of intuitively understand and predict how materials deform and move, a scenario that has been frequently seen in Science-Fiction movies.

cs.CE↗