24 papers · ranked by Valyu relevance
Authors not listed
Inverse molecular design aims to generate novel chemical structures that satisfy multiple property constraints, yet reinforcement-learning (RL) fine-tuning can be sensitive to how objectives are converted into a scalar reward. Here, we systematically analyze how scalarization choices and stabilization mechanisms shape…
Chen Wang, Zhaochun Li, Bai, Jionghao + 4 more
Reinforcement Learning (RL) is essential for enhancing the reasoning capabilities of large language models (LLMs), yet the widely adopted Group Relative Policy Optimization (GRPO) suffers from entropy collapse, causing exploration to vanish and policies to converge prematurely. As a result, RL is widely believed to be…
Giorgio Taricco, Miguel Rubi
Entropy regularization is a recurring mechanism in reinforcement learning (RL), but its meaning changes across algorithmic settings. In classical online RL, entropy encourages exploration and smooths policy improvement; in inverse RL and imitation learning, maximum-entropy resolves ambiguity among expert-consistent…
Saira Jabeen, Asa Ben-Hur
The successful use of deep learning in computational biology depends on the ability to extract meaningful biological information from the trained models. Recent work has demonstrated that the attention maps generated by self-attention layers can be interpreted to predict cooperativity between binding of transcription…
Andrzej Cichocki, Toshihisa Tanaka, Frank Nielsen, Sergio Cruces + 2 more
This paper introduces a broad class of Mirror Descent (MD) and Generalized Exponentiated Gradient (GEG) algorithms derived from trace-form entropies defined via deformed logarithms. Leveraging these generalized entropies yields MD and GEG algorithms with improved convergence behavior, robustness against vanishing and…
Jingwei Chen
Self-referential learning—training a model on data it generated itself—promises boundless scalability but chronically suffers from model collapse: language models degenerate into repetitive text, GANs drop modes, and reinforcement-learning policies over-exploit. Although practitioners employ ad hoc fixes such as…
Kleyton da Costa, Bernardo Modenesi
Tsallis statistics generalizes Boltzmann-Gibbs statistical mechanics through a single real parameter $q$ that controls the weight assigned to rare and frequent events. Originally proposed to describe physical systems with long-range correlations, multifractal geometry, and heavy-tailed fluctuations, the framework has…
Di Carlo, Luca, Goddard, Chase + 2 more
Modern neural networks exhibit a striking property: basins of attraction in the loss landscape are often connected by low-loss paths, yet optimization dynamics generally remain confined to a single convex basin (Baity-Jesi et al., 2019; Juneja et al., 2023) and rarely explore intermediate points. We resolve this…
Thi Lich Nghiem, Pierre Maréchal, Nikolai Leonenko
We study Maximum Entropy density estimation on continuous domains under finitely many moment constraints, formulated as the minimization of the Kullback-Leibler divergence with respect to a reference measure. To model uncertainty in empirical moments, constraints are relaxed through convex penalty functions, leading to…
Qiaozhe Zhang, Jun Sun, Ruijie Zhang, Yingzhuang Liu
Sharpness (of the loss minima) is a common measure to investigate the generalization of neural networks. Intuitively speaking, the flatter the landscape near the minima is, the better generalization might be. Unfortunately, the correlation between many existing sharpness measures and the generalization is usually not…
Authors not listed
Gibbs’ paradox—the apparent discontinuity in mixing entropy for gases of varying similarity and the seeming reversibility of mixing-separation cycles—has resisted fully satisfactory resolution for 150 years. We present a solution based on categorical state theory, which posits that physical con f igurations are…
Wang, Jingyuan, Ji, Jiahao
This article serves as the regression analysis lecture notes in the Intelligent Computing course cluster (including the courses of Artificial Intelligence, Data Mining, Machine Learning, and Pattern Recognition) at the School of Computer Science and Engineering, Beihang University. It aims to provide students – who are…
Antony Mizzi, David M. Walker, Michael Small, José F. F. Mendes
We derive a penalty strength criterion for ridge regression using stochastic complexity, which is a refined variant of the minimum description length principle. Since stochastic complexity does not typically account for the effect of regularization on complexity, despite its ability to simplify models, we are required…
Louella Seo, Ian Farran, Ahmed Aslam, Xinyun Li + 4 more
Structure-based drug design has traditionally focused on optimizing static, enthalpic interactions between ligands and proteins or on displacing binding site solvent molecules to entropically favor binding. A potentially large contributor to binding thermodynamics is the difference in conformational entropy of the…
Charlotte A. Miller, Stephanie A. Wankowicz
Protein-ligand binding is governed by free energy, comprising both enthalpic and entropic contributions. Yet structural interpretations of binding thermodynamics have predominantly focused on enthalpic interactions, largely neglecting entropy because it is difficult to quantify from static structural models. Here, we…
Laurent Caraffa
Modern machine learning relies on a collection of empirically successful but theoretically heterogeneous regularization techniques, such as weight decay, dropout, and exponential moving averages. At the same time, the rapidly increasing energetic cost of training large models raises the question of whether learning…
John Knight
Latent Factor Analysis via Dynamical Systems (LFADS) is a powerful variational autoencoder for inferring neural population dynamics from spike train data. However, LFADS suffers from pos-terior collapse, where the learned posterior collapses to the prior, eliminating meaningful latent representations. Current solutions…
Authors not listed
Collective variables (CVs) are essential for interpreting and accelerating rare events in molecular simulations. However, their design remains limited by the requirement of differentiability with respect to atomic coordinates. This constraint excludes many powerful structural descriptors that are routinely used for…
Guanghui Zhang, Xinran Wang, Steven J. Luck
Regularization has been extensively used in multivariate pattern classification (MVPA; decoding) of EEG data to mitigate the risk of overfitting. N-fold cross-validation is also used to mitigate this risk, and it is often combined with averaging across trials to improve the signal-to-noise ratio. However, the impact of…
Benjamin Kuznets-Speck, Jaekwon Jung, Pornchanan Pholraksa, Adrianne Zhong + 4 more
Classification and regression are cornerstones of computational biology and science at large, from identifying cell types to stratifying patients by disease state. Current deep learning classifiers provide accurate predictions but offer neither uncertainty estimates nor insight into which features matter most. On the…
Yuhua Fan, Ilkka Launonen, Mikko J Sillanpää, Patrik Waldmann
High-dimensional genomic datasets contain complex patterns shaped by substantial biological noise, which pose major challenges for predictive modeling in genetics and breeding. Residual neural networks (ResNets) provide a powerful framework for capturing nonlinear genomic effects, but often overfit in settings where…
Authors not listed
Machine learning models are increasingly applied to heterogeneous materials datasets spanning different synthesis routes, measurement protocols, and structural classes. Although multi-task and representation-learning approaches are commonly used to improve predictive performance, the latent representations learned by…
Authors not listed
Phase equilibrium calculations are crucial in chemical engineering design and optimization processes. The PC-SAFT equation of state (EoS) can precisely calculate phase equilibrium, but is relatively complex and computationally intensive. Surrogate models are mathematically simple models that map or regress the…
Ray Zirui Zhang, Christopher E. Miles, Xiaohui Xie, John S. Lowengrub
We propose a new neural network based method for solving inverse problems for partial differential equations (PDEs) by formulating the PDE inverse problem as a bilevel optimization problem. At the upper level, we minimize the data loss with respect to the PDE parameters. At the lower level, we train a neural network to…