arXiv subjects

Rohan Shah

Publications and source records attributed to Rohan Shah.

9 recordsLinked to original sources

arXiv2026

Spatiotemporal Disk Packing for Directed Growth of Complex Geometries

Growing complex shapes requires control over both where growth begins and how it evolves in time. Here, we introduce a geometric framework for growing prescribed 2D shapes using a disk packing algorithm. In this approach, a target geometry is filled by disks whose centers define where growth is initiated and whose radii define how long each region is allowed to grow. The allowed disk sizes are constrained by the physics of the process, including the growth velocity, the time required to initiate each growth event, and the number of initiations that can occur in parallel. To generate physically realizable packings, we introduce the Largest Gap Algorithm (LGA), which sequentially fills the largest remaining gaps in a target shape with the largest disk that satisfies both geometric and kinetic constraints. We show that this method produces high coverage packings for a variety of geometries and that the resulting packings can be directly converted into spatiotemporal packing instructions. We then demonstrate that these instructions can be realized experimentally using multi-point initiation of frontal polymerization in viscosified dicyclopentadiene (DCPD) resin using CO$_2$ laser. Our results show that complex shapes can be grown by programming a small number of local initiation events, providing a simple connection between geometry and dynamics of growth.

cond-mat.soft↗

arXiv2025

Active contacts create controllable friction

Sliding friction between two dry surfaces is reasonably described by the speed-independent Amonton-Coulomb friction force law. However, there are many situations where the frictional contact points between two surfaces are "active" and may not all be moving at the same relative speed. In this work we study the sliding friction properties of a system with multiple active contacts each with independent and controllable speed. We demonstrate that multiple active contacts can produce controllable speed-dependent sliding friction forces, despite each individual contact exhibiting a speed-independent friction. We study in experiment a rotating carousel with ten speed-controlled wheels in frictional contact with the ground. We first vary the contact speeds and demonstrate that the equilibrium system speed is the median of the active contact speeds. Next we directly measure the ground reaction forces and observe how the contact speeds can control the force-speed curve of the system. In the final experiments we demonstrate how control of the force-speed curve can create sliding friction with a controllable effective viscosity and controllable sliding friction coefficient. Surprisingly, we are able to demonstrate that frictional contacts can create near frictionless sliding with appropriate force-speed control. By revealing how active contacts can shape the force-speed behavior of dry sliding friction systems we can better understand animal and robot locomotion, and furthermore open up opportunities for new engineered surfaces to control sliding friction.

cond-mat.soft↗

arXiv2024

Introduction to Martingales

This paper introduces Martingales by covering introductory measure theory concepts and the Lebesgue Integration and Conditional Expectation. It follows up with proofs of Kolomorgov's Theorem on conditional expectations, the Martingale Property, and the Pythagorean Theorem on Martingales. Finally, it ends with Martingales' applications in finance.

math.PR↗

arXiv2023

Scaling transformer neural networks for skillful and reliable medium-range weather forecasting

Weather forecasting is a fundamental problem for anticipating and mitigating the impacts of climate change. Recently, data-driven approaches for weather forecasting based on deep learning have shown great promise, achieving accuracies that are competitive with operational systems. However, those methods often employ complex, customized architectures without sufficient ablation analysis, making it difficult to understand what truly contributes to their success. Here we introduce Stormer, a simple transformer model that achieves state-of-the-art performance on weather forecasting with minimal changes to the standard transformer backbone. We identify the key components of Stormer through careful empirical analyses, including weather-specific embedding, randomized dynamics forecast, and pressure-weighted loss. At the core of Stormer is a randomized forecasting objective that trains the model to forecast the weather dynamics over varying time intervals. During inference, this allows us to produce multiple forecasts for a target lead time and combine them to obtain better forecast accuracy. On WeatherBench 2, Stormer performs competitively at short to medium-range forecasts and outperforms current methods beyond 7 days, while requiring orders-of-magnitude less training data and compute. Additionally, we demonstrate Stormer's favorable scaling properties, showing consistent improvements in forecast accuracy with increases in model size and training tokens. Code and checkpoints are available at https://github.com/tung-nd/stormer.

physics.ao-ph↗

arXiv2023

Turn Down the Noise: Leveraging Diffusion Models for Test-time Adaptation via Pseudo-label Ensembling

The goal of test-time adaptation is to adapt a source-pretrained model to a continuously changing target domain without relying on any source data. Typically, this is either done by updating the parameters of the model (model adaptation) using inputs from the target domain or by modifying the inputs themselves (input adaptation). However, methods that modify the model suffer from the issue of compounding noisy updates whereas methods that modify the input need to adapt to every new data point from scratch while also struggling with certain domain shifts. We introduce an approach that leverages a pre-trained diffusion model to project the target domain images closer to the source domain and iteratively updates the model via pseudo-label ensembling. Our method combines the advantages of model and input adaptations while mitigating their shortcomings. Our experiments on CIFAR-10C demonstrate the superiority of our approach, outperforming the strongest baseline by an average of 1.7% across 15 diverse corruptions and surpassing the strongest input adaptation baseline by an average of 18%.

cs.CV↗

arXiv2021

PAC Mode Estimation using PPR Martingale Confidence Sequences

We consider the problem of correctly identifying the \textit{mode} of a discrete distribution $\mathcal{P}$ with sufficiently high probability by observing a sequence of i.i.d. samples drawn from $\mathcal{P}$. This problem reduces to the estimation of a single parameter when $\mathcal{P}$ has a support set of size $K = 2$. After noting that this special case is tackled very well by prior-posterior-ratio (PPR) martingale confidence sequences \citep{waudby-ramdas-ppr}, we propose a generalisation to mode estimation, in which $\mathcal{P}$ may take $K \geq 2$ values. To begin, we show that the "one-versus-one" principle to generalise from $K = 2$ to $K \geq 2$ classes is more efficient than the "one-versus-rest" alternative. We then prove that our resulting stopping rule, denoted PPR-1v1, is asymptotically optimal (as the mistake probability is taken to $0$). PPR-1v1 is parameter-free and computationally light, and incurs significantly fewer samples than competitors even in the non-asymptotic regime. We demonstrate its gains in two practical applications of sampling: election forecasting and verification of smart contracts in blockchains.

stat.ME↗

arXiv2018

ExoMol molecular line lists - XXVI: spectra of SH and NS

Line lists for the sulphur-containing molecules SH (the mercapto radical) and NS are computed as part of the ExoMol project. These line lists consider transitions within the $X$ ${}^2\Pi$ ground state for $^{32}$SH, $^{33}$SH, $^{34}$SH and $^{\text{32}}$SD, and $^{14}$N$^{32}$S, $^{14}$N$^{33}$S, $^{14}$N$^{34}$S, $^{14}$N$^{36}$S and $^{15}$N$^{32}$S. Ab initio potential energy (PEC) and spin-orbit coupling (SOC) curves are computed and then improved by fitting to experimentally observed transitions. Fully ab initio dipole moment curves (DMCs) computed at high level of theory are used to produce the final line lists. For SH, our fit gives a root-mean-square (rms) error of 0.03 cm$^{-1}$ between the observed ($v_{\rm max}=4$, $J_{\rm max} = 34.5$) and calculated transitions wavenumbers; this is extrapolated such that all $X$ $^2\Pi$ rotational-vibrational-electronic (rovibronic) bound states are considered. For $^{\text{32}}$SH the resulting line list contains about 81000 transitions and 2300 rovibronic states, considering levels up to $v_{\rm max} = 14$ and $J_{\rm max} = 60.5$. For NS the refinement used a combination of experimentally determined frequencies and energy levels and led to an rms fitting error of 0.002 cm$^{-1}$. Each NS calculated line list includes around 2.8 million transitions and 31000 rovibronic states with a vibrational range up to $v=53$ and rotational range to $J=235.5$, which covers up to 23000 cm$^{-1}$. Both line lists should be complete for temperatures up to 5000 K. Example spectra simulated using this line list are shown and comparisons made to the existing data in the CDMS database. The line lists are available from the CDS (http://cdsarc.u-strasbg.fr) and ExoMol (www.exomol.com) data bases.

astro-ph.SR↗

arXiv2016

Using Artificial Intelligence to Identify State Secrets

Whether officials can be trusted to protect national security information has become a matter of great public controversy, reigniting a long-standing debate about the scope and nature of official secrecy. The declassification of millions of electronic records has made it possible to analyze these issues with greater rigor and precision. Using machine-learning methods, we examined nearly a million State Department cables from the 1970s to identify features of records that are more likely to be classified, such as international negotiations, military operations, and high-level communications. Even with incomplete data, algorithms can use such features to identify 90% of classified cables with <11% false positives. But our results also show that there are longstanding problems in the identification of sensitive information. Error analysis reveals many examples of both overclassification and underclassification. This indicates both the need for research on inter-coder reliability among officials as to what constitutes classified material and the opportunity to develop recommender systems to better manage both classification and declassification.

cs.CY↗

arXiv2015

Estimating Residual Connectivity for Random Graphs

Computation of the probability that a random graph is connected is a challenging problem, so it is natural to turn to approximations such as Monte Carlo methods. We describe sequential importance resampling and splitting algorithms for the estimation of these probabilities. The importance sampling steps of these algorithms involve identifying vertices that must be present in order for the random graph to be connected, and conditioning on the corresponding events. We provide numerical results demonstrating the effectiveness of the proposed algorithm.

stat.CO↗