Search for: All records

In this paper, we present a generalized, data-driven collisional operator for one-component plasmas, learned from molecular dynamics simulations, to extend the collisional kinetic model beyond the weakly coupled regime. The proposed operator features an anisotropic, non-stationary collision kernel that accounts for particle correlations typically neglected in classical Landau formulations. To enable efficient numerical evaluation, we develop a fast spectral separation method that represents the kernel as a low-rank tensor product of univariate basis functions. This formulation admits an O(N log N) algorithm via fast Fourier transforms and preserves key physical properties, including discrete conservation laws and the H-theorem, through a structure-preserving central difference discretization. Numerical experiments demonstrate that the proposed model accurately captures plasma dynamics in the moderately coupled regime beyond the standard Landau model while maintaining high computational efficiency and structure-preserving properties. 

As electric vehicles (EVs) gradually replace fuel vehicles and provide transportation services in cities, e.g., electric taxi fleets, solar-powered charging stations with energy storage systems have been deployed to provide charging services for EV fleets. The mixture of solar-powered and traditional charging stations brings efficiency challenges to charging stations and reliability challenges to power systems. In this article, we explore e-taxis’ mobility and charging demand flexibility to co-optimize service quality of e-taxi fleets and system cost of charging infrastructures, e.g., solar power under-utilization and reliability issues of power distribution networks due to reverse power flow. We propose SAC, an e-taxi coordination framework to dispatch e-taxis for charging or serving passengers under spatial-temporal dynamics of renewable energy and passenger mobility, which integrates the renewable power generation estimation from a forecast system. Moreover, we extend our design to a stochastic Model Predictive Control problem to handle the uncertainty of solar power generation, aiming to fully utilize generated solar power. Our data-driven evaluation shows that SAC significantly outperforms existing solutions, enhancing the usage rate of solar power by up to 172.6%, while maintaining e-taxi service quality with very small overhead, i.e., reducing the supply-demand ratio by 2.2%. 

We introduce a data-driven approach to learn generalized collision operators from molecular dynamics. Unlike conventional models (e.g., Landau), the present operator takes a symmetry-breaking form that depends not only on the relative velocity but also on the average velocity of the collision pair, capturing heterogeneous energy transfer arising from collective interactions with the environment. Numerical results show that preserving this overlooked anisotropy is crucial for predicting plasma kinetics and transport coefficients in the moderate coupling regime, where the Landau model fails. 

Accelerating the solution process of preventive Security Constrained Optimal Power Flow (SCOPF) is studied. In an iterative algorithmic framework, knowledge of the relatively sparse critical contingencies can greatly reduce the problem size and hence solution time. Predictors that can predict the critical contingencies that must be included in the SCOPF formulation are trained and integrated into an iterative algorithm. As different types of prediction errors — false negatives and false positives — have markedly different impact on algorithm solution time, a novel time-aware loss function is designed and calibrated for training predictors that directly minimizes algorithm run-time. A multi-objective loss function that incorporates both this time-aware loss and the accuracy promoting binary cross entropy (BCE) loss is then designed and tuned. Comprehensive evaluation of the time-aware predictor-assisted iterative algorithm is conducted based on the IEEE 118-bus system. Effective re-balancing of false negatives and false positives is observed, and significant reduction of algorithm run-time is achieved with the developed time-aware predictors. 

We introduce Quantized Language-Image Pretraining (QLIP), a visual tokenization method that combines state-of-the-art reconstruction quality with state-of-the-art zero-shot image understanding. QLIP trains a binary-spherical-quantization-based autoencoder with reconstruction and language-image alignment objectives. We are the first to show that the two objectives do not need to be at odds. We balance the two loss terms dynamically during training and show that a two-stage training pipeline effectively mixes the large-batch requirements of image-language pre-training with the memory bottleneck imposed by the reconstruction objective. We validate the effectiveness of QLIP for multimodal understanding and text-conditioned image generation with a single model. Specifically, QLIP serves as a drop-in replacement for the visual encoder for LLaVA and the image tokenizer for LlamaGen with comparable or even better performance. Finally, we demonstrate that QLIP enables a unified mixed-modality auto-regressive model for understanding and generation. 

Abstract Protein language models, like the popular ESM2, are widely used tools for extracting evolution-based protein representations and have achieved significant success on downstream biological tasks. Representations based on sequence and structure models, however, show significant performance differences depending on the downstream task. A major open problem is to obtain representations that best capture both the evolutionary and structural properties of proteins in general. Here we introduceImplicitStructureModel(ISM), a sequence-only input model with structurally-enriched representations that outperforms state-of-the-art sequence models on several well-studied benchmarks including mutation stability assessment and structure prediction. Our key innovations are a microenvironment-based autoencoder for generating structure tokens and a self-supervised training objective that distills these tokens into ESM2’s pre-trained model. We have madeISM’s structure-enriched weights easily available: integrating ISM into any application using ESM2 requires changing only a single line of code. Our code is available athttps://github.com/jozhang97/ISM.