arXiv subjects

Aditya Gupta

Publications and source records attributed to Aditya Gupta.

At least 19 recordsLinked to original sources

arXiv2026

Learning Hierarchical Patch Splitting Policies for Faster Vision Transformers

Vision Transformers (ViTs) are bottlenecked by the length of their input token sequence. Standard patchification fixes this length with a single patch size applied across the image, regardless of regional content. Prior work on adaptive patch sizing reduces token count by assigning coarser patches to visually simple regions, but relies on noisy pixel-level heuristics such as local entropy to make patchification decisions. We argue that this compute allocation should be grounded in the model's interpretation of task relevant information, not in raw pixel statistics. We introduce Split Policy Learned Image Tokenization which learns to allocate token resolution based on what the model finds meaningful, producing shorter token sequences that are better matched to task-relevant content. SPLIT matches the accuracy of full-resolution ViT classification baselines while using 30% fewer tokens. The same gains hold on dense prediction: SPLIT achieves competitive accuracy on COCO and ADE20K, demonstrating that semantically aligned patchification generalises well beyond classification.

cs.CV↗

arXiv2026

Modular Constructions of g-Golomb Rulers

A set $\mathcal{G}$ of integers is a $g$-Golomb ruler if each positive difference appears at most $g$ times between any 2 elements of the set, and $G(g,n)$ denotes the minimum diameter of such a ruler with $n$ marks. We prove a general lemma for passing from certain modular constructions to ordinary $g$-Golomb rulers. The key point is that, in a modular $g$-Golomb ruler, no cyclic gap length can occur more than $g$ times. This gives a larger guaranteed cut than the previous average gap argument. We apply this lemma to cyclic relative difference sets, Singer sets, Ruzsa--Spence rulers, and Paley quadratic residues to provide many competing constructions for $g$-Golomb Rulers. A computation on the grid $1\le g\le500$, $n=g+b$, $2\le b\le500$, compares the four resulting construction families.

math.CO↗

arXiv2026

PhySimCode: A Benchmark and Evaluation Method for Physics Video to Code Generation

Multimodal large language models (MLLMs) are good at video understanding and code generation, but it remains unclear whether they can recover the physical laws underlying videos, a capability essential for converting videos into executable simulations and for providing structured physical supervision in robotics. Captioning motion video does not imply that a model has inferred its governing dynamics, estimated quantitative parameters, or can regenerate the observed scene from code. Existing video-physics benchmarks evaluate plausibility, property estimation, or future-state prediction, but offer limited insight into whether models can translate visual dynamics into runnable simulation code. We introduce PhySimCode, a benchmark for physics video-to-code generation: given only a raw video, a model must infer the relevant physics, recover parameter values, and generate runnable Python that reproduces the input video. The dataset contains 160.6K procedurally generated samples spanning 162 physics phenomena across analytic SciPy simulations and PyBullet-rendered rigid-body, articulated, fluid, and contact-rich scenes; each sample bundles the simulation code, physical parameters, rendered video, and a structured chain-of-thought covering observations, laws, ODEs, numerical methods, and validation checks. We evaluate ten state-of-the-art closed- and open-source MLLMs along physical-law correctness, simulation equivalence, code executability, and parameter recovery. Models reliably state physics laws (4.27-4.93/5) yet score only 1.51-2.47/5 on inferring them from video, and the best system (Claude Sonnet 4.6) recovers just 7.7% of ground-truth parameters within $\pm 20\%$, exposing a substantial video-to-simulation gap. The 2,430-video evaluation set is available at https://huggingface.co/datasets/SourajaKundu123/PhySimCode and the evaluation code at https://github.com/SourajaKundu/PhySimCode

cs.CV↗

arXiv2026

Novel View Synthesis using DDIM Inversion

Single-image novel view synthesis requires viewpoint-dependent structural changes while retaining source-scene information and inferring unobserved content. We formulate this task as a camera-conditioned initialization problem for a frozen pretrained latent diffusion model. At an intermediate DDIM inversion timestep, we decompose the source representation into a structural mean and a source noise residual. A pose-conditioned translation U-Net (TUNet) learns to predict the target-view mean from the source mean. We then introduce a score-guided fusion strategy: the predicted mean and source noise residual form a proposal that is evaluated by the frozen denoiser, whose noise prediction is combined with the translated mean to initialize DDIM sampling. Only TUNet is trained, while the pretrained model synthesizes the final view without explicit reconstruction or rendering of a shared 3D scene. Experiments on MVImgNet and RealEstate10K demonstrate competitive performance. Beyond image-level metrics, Wasserstein comparisons show that the corrected initialization yields trajectory-curvature distributions closer to those of ground-truth target-initialized sampling than raw fusion. These results support structural translation and source-conditioned initialization as a practical approach to reusing frozen diffusion priors for novel view synthesis.

cs.CV↗

arXiv2026

Deep Reinforcement Learning for Reliability Based Bi-Objective Portfolio Optimization

Portfolio optimization under uncertainty is inherently a multi-objective decision problem involving complex interactions among return, risk, market dynamics, and practical investment constraints. Existing reliability based portfolio optimization approaches primarily rely on static optimization frameworks and often fail to capture sequential decision making, tail risk, and market frictions such as transaction costs. To address these limitations, we propose a deep reinforcement learning framework for multi-objective reliability based portfolio optimization (MORP-DRL). The proposed framework jointly optimizes expected return and downside risk using three complementary risk measures: variance, Conditional Value-at-Risk (CVaR), and Entropic Value-at-Risk (EVaR). To model uncertainty and heavy-tailed market behavior, asset returns are represented using GARCH(1,1), Extreme Value Theory, and a t-copula dependence structure, while realistic scenarios are generated through quasi-Monte Carlo simulation. A Proximal Policy Optimization (PPO) based strategy is developed under practical constraints including transaction costs and portfolio bounds, and is benchmarked against NSGA-II. Experiments on ten global equity indices across pre-COVID, COVID, and post-COVID market regimes demonstrate that MORP-DRL achieves competitive risk-return performance, reduced downside risk during periods of market stress, and scalability to high-dimensional portfolio settings.

cs.LG↗

arXiv2026

Perpetual Futures for Stocks: The SpaceX Pre-IPO Market

Robert Shiller proposed perpetual futures in 1993 to create derivative markets for assets that are illiquid or whose price cannot be observed directly, such as single family homes, human capital, and the consumer price index. The crypto markets later built the instrument for a different reason and with a different funding rule. We give a single no arbitrage result that nests both designs: the perpetual price is the present value of a benchmark flow discounted at the funding rate, so the funding rule chooses both the benchmark and the discount. We give a random time change representation in which the price is the expected spot at the first event of a clock whose intensity is the funding rate, use it to show that stochastic volatility moves the basis only through the carry, so a volatility risk premium and not volatility itself can break the peg, read price discovery as the convergence of a Doob martingale driven by a stochastic approximation, and give a segmented market equilibrium that makes the pre listing premium structural rather than behavioural. The June 2026 SpaceX pre-IPO market is the first large scale realization of the idea for an equity claim, and we find that the perpetual consensus forecast the secondary clearing price more accurately than the bookbuilt offer. We close with other equity applications.

q-fin.PR↗

arXiv2026

Optimal Diameters of High Multiplicity g-Golomb Rulers

A set $\mathcal{G}$ of integers is called a $g$-Golomb ruler of length $n$ if the difference between any two distinct elements of $\mathcal{G}$ is repeated at most $g$ times. If $g=1$, these are also called $B_2$-sets, Sidon sets, and Babcock sets. We define $G(g,n)$ to represent the minimum diameter of a $g$-Golomb Ruler. In this paper, we prove that for all $b\ge 1$, if $g \ge \frac{7}{4}\left(b^{3/2} -b\right)+1,$ then $G(g,g+b)=g+2b-2$. Sharper bounds are given for $b\le 18$. The main technique is through an arithmetic property of the integers that are \emph{not} in a $g$-Golomb ruler, leading us to introduce LM rulers, a new class of rulers where every distance $d$ occurs as a difference at most $d-1$ times. We show that the minimum diameter of an $n$-element LM ruler $L(n)$ is $\sqrt{8/9} \cdot (n-1)^{3/2} \le L(n) \le \frac{7}{4}\left((n+1)^{3/2}-(n+1)\right).$

math.CO↗

arXiv2026

PROPER Agents: Proactivity Driven Personalized Agents for Advancing Knowledge Gap Navigation

Current approaches to proactive assistance move beyond the ask-and-respond paradigm by anticipating user needs. In practice, they either burden users with clarifying questions or rely on context-based extrapolation, often leading to unnecessary or mistimed interventions. Such systems lack explicit mechanisms to model users' knowledge gaps, resulting in incomplete or suboptimal task outcomes. To address this, we propose PROPER, a framework that explicitly models user-specific knowledge gaps in a controlled manner. Central to our approach is the notion of dimensions: structured, task-relevant factors that define the considerations required for effective task completion. Given a user query, the DGA (Dimension Generating Agent) identifies explicit dimensions (from the user's query) and generates a set of candidate implicit dimensions capturing unarticulated aspects of the task. The RGA (Response Generating Agent) integrates both explicit and implicit dimensions selectively to produce personalized, context-aware, and proactively informative responses. We evaluate PROPER across multiple domains using a structured, gap-aware rubric that measures coverage, initiative appropriateness, and intent alignment. PROPER improves on quality scores and win rates across all domains, achieving up to 84% gains in single-turn evaluation and consistent dominance in multi-turn interactions. All code for PROPER is available at: https://github.com/i-kiran/ProPer-Agent.

cs.LG↗

arXiv2026

XOXO: Stealthy Cross-Origin Context Poisoning Attacks against AI Coding Assistants

AI coding assistants are widely used for tasks like code generation. These tools now require large and complex contexts, automatically sourced from various origins$\unicode{x2014}$across files, projects, and contributors$\unicode{x2014}$forming part of the prompt fed to underlying LLMs. This automatic context-gathering introduces new vulnerabilities, allowing attackers to subtly poison input to compromise the assistant's outputs, potentially generating vulnerable code or introducing critical errors. We propose a novel attack, Cross-Origin Context Poisoning (XOXO), that is challenging to detect as it relies on adversarial code modifications that are semantically equivalent. Traditional program analysis techniques struggle to identify these perturbations since the semantics of the code remains correct, making it appear legitimate. This allows attackers to manipulate coding assistants into producing incorrect outputs, while shifting the blame to the victim developer. We introduce a novel, task-agnostic, black-box attack algorithm GCGS that systematically searches the transformation space using a Cayley Graph, achieving a 75.72% attack success rate on average across five tasks and eleven models, including GPT 4.1 and Claude 3.5 Sonnet v2 used by popular AI coding assistants. Furthermore, defenses like adversarial fine-tuning are ineffective against our attack, underscoring the need for new security measures in LLM-powered coding tools.

cs.CR↗

arXiv2026

Transient Thermodynamic Efficiency of Adaptive Inference in Continuously Nonstationary Environments

Adaptive physical and biological systems continually process fluctuating information from their environments. When the environment is nonstationary, inference itself becomes a nonequilibrium process with thermodynamic cost. We analyse a minimal stochastic model which is an overdamped particle in an adaptive double well potential whose control parameter tracks a drifting Ornstein Uhlenbeck signal. Using stochastic energetics, we derive explicit expressions for entropy production, mutual information rate, and a time dependent learning efficiency. High precision Langevin simulations reveal transient peaks in learning efficiency during rapid environmental shifts, absent in steady state averages. These results identify transient adaptive regimes as moments of maximal information to energy conversion, highlighting that maximal thermodynamic learning performance arises transiently rather than in steady state. Throughout this work, the environment is treated as an externally driven stochastic signal rather than a thermodynamic subsystem under control, and its intrinsic entropy production is therefore excluded from the thermodynamic accounting.

cond-mat.stat-mech↗

arXiv2025

QLENS: Towards A Quantum Perspective of Language Transformers

In natural language processing, current methods for understanding Transformers are successful at identifying intermediate predictions during a model's inference. However, these approaches function as limited diagnostic checkpoints, lacking a mathematical framework for mechanistically modeling how each layer facilitates transitions between these evolving states. This interpretability gap and past successes of interdisciplinary outlooks inspire us to turn to physics in search of a descriptive mathematical framework for Transformers. We observe that language models are intrinsically probabilistic, an attribute that is echoed in the core postulates of quantum mechanics. This parallel inspires us to translate insights from this discipline to that of natural language processing. Towards this objective, we propose QLENS a novel attempt to develop a physics-based perspective on the Transformer generation process. Under QLENS, a Transformer is studied by converting its latent activations into a state vector in a Hilbert space derived from the model's output units. This state subsequently evolves through hidden layers - reformulated as unitary operators and analogously defined Hamiltonians - during inference. The model's final probability distribution is obtained by applying the Born rule to the end state using a specific measurement operator. To demonstrate QLENS's potential, we conduct a proof-of-concept by probing a toy Transformer to investigate the influence of individual layers in a model's prediction trajectory. We present our work as a foundation for cross-domain insights to be leveraged towards a broader understanding of Transformers.

cs.LG↗

arXiv2025

NordFKB: a fine-grained benchmark dataset for geospatial AI in Norway

We present NordFKB, a fine-grained benchmark dataset for geospatial AI in Norway, derived from the authoritative, highly accurate, national Felles KartdataBase (FKB). The dataset contains high-resolution orthophotos paired with detailed annotations for 36 semantic classes, including both per-class binary segmentation masks in GeoTIFF format and COCO-style bounding box annotations. Data is collected from seven geographically diverse areas, ensuring variation in climate, topography, and urbanization. Only tiles containing at least one annotated object are included, and training/validation splits are created through random sampling across areas to ensure representative class and context distributions. Human expert review and quality control ensures high annotation accuracy. Alongside the dataset, we release a benchmarking repository with standardized evaluation protocols and tools for semantic segmentation and object detection, enabling reproducible and comparable research. NordFKB provides a robust foundation for advancing AI methods in mapping, land administration, and spatial planning, and paves the way for future expansions in coverage, temporal scope, and data modalities.

cs.CV↗

arXiv2025

Advancing Digital Accessibility In Digital Pharmacy, Healthcare, And Wearable Devices: Inclusive Solutions for Enhanced Patient Engagement

Modern healthcare facilities demand digital accessibility to guarantee equal access to telemedicine platforms, online pharmacy services, and health monitoring devices that can be worn or are handy. With the rising call for the implementation of robust digital healthcare solutions, people with disabilities encounter impediments in their endeavor of managing and getting accustomed to these modern technologies owing to insufficient accessibility features. The paper highlights the role of comprehensive solutions for enhanced patient engagement and usability, particularly, in digital pharmacy, healthcare, and wearable devices. Besides, it elucidates the key obstructions faced by users experiencing auditory, visual, cognitive, and motor impairments. Through a kind consideration of present accessibility guidelines, practices, and emerging technologies, the paper provides a holistic overview by offering innovative solutions, accentuating the vitality of compliance with Web Content Accessibility Guidelines (WCAG), Americans with Disabilities Act (ADA), and other regulatory structures to foster easy access to digital healthcare services. Moreover, there is due focus on using AI-driven tools, speech-activated interfaces, and tactile feedback in wearable health devices to assist persons with disabilities. The outcome of the research explicates the necessity of prioritizing accessibility for individuals with disabilities and cultivating a culture where healthcare providers, policymakers, and officials build a patient-centered digital healthcare ecosystem that is all-encompassing in nature.

cs.HC↗

arXiv2025

Advancing Digital Accessibility: Integrating AR/VR and Health Tech for Inclusive Healthcare Solutions

Modern healthcare domain incorporates a feature of digital accessibility to ensure seamless flow of online services for the patients. However, this feature of digital accessibility poses a challenge particularly for patients with disabilities. To eradicate this issue and provide immersive and user-friendly experiences, evolving technologies like Augmented Reality (AR) and Virtual Reality (VR) are integrated in medical applications to enhance accessibility. The present research paper aims to study inclusivity and accessibility features of AR/VR in revolutionizing healthcare practices especially in domains like telemedicine, patient education, assistive tools, and rehabilitation for persons with disabilities. The current trends of advancements and case studies are also analyzed to measure the efficacy of AR/VR in healthcare. Moreover, the paper entails a detailed analysis of the challenges of its adoption particularly technical limitations, implementation costs, and regulatory aspects. Finally, the paper concludes with recommendations for integrating AR/VR to foster a more equitable and inclusive healthcare system and provide individuals with auditory, visual, and motor impairments with digital healthcare solutions.

cs.HC↗

arXiv2025

Bridging the Gap: Enhancing Digital Accessibility for Medicaid Populations in Telehealth Adoption

The swift evolution of telehealth has revolutionized how medical professionals deliver healthcare services and boost convenience and accessibility. Yet, the Medicaid population encounters several impediments in utilizing facilities especially owing to poor internet connectivity, less awareness about digital platforms, and a shortage of assistive technologies. The paper aims to explicate key factors behind digital accessibility for Medicaid populations and expounds robust solutions to eradicate these challenges. Through inclusive design ideas, AI-assisted technologies, and all-encompassing policies by the concerned authorities, healthcare professionals can enhance usability and efficacy and thus better serve the needy. This revolution not only enhances convenience but also expands access, mainly for underserved groups such as rural populations or those with mobility issues, thereby ensuring inclusivity and flexibility in the healthcare domain. Besides, the paper highlights the vitality of collaboration between healthcare professionals, policymakers, and tech developers in unveiling the accessibility and usability impediments. What else helps in minimizing healthcare differences and enhancing patient outcomes is guaranteeing equitable access to telehealth for Medicaid beneficiaries. The paper systematically offers major recommendations to increase digital accessibility in telehealth, thereby creating a patient-oriented and all-encompassing healthcare system.

cs.CY↗

arXiv2025

A New HOPE: Domain-agnostic Automatic Evaluation of Text Chunking

Document chunking fundamentally impacts Retrieval-Augmented Generation (RAG) by determining how source materials are segmented before indexing. Despite evidence that Large Language Models (LLMs) are sensitive to the layout and structure of retrieved data, there is currently no framework to analyze the impact of different chunking methods. In this paper, we introduce a novel methodology that defines essential characteristics of the chunking process at three levels: intrinsic passage properties, extrinsic passage properties, and passages-document coherence. We propose HOPE (Holistic Passage Evaluation), a domain-agnostic, automatic evaluation metric that quantifies and aggregates these characteristics. Our empirical evaluations across seven domains demonstrate that the HOPE metric correlates significantly (p > 0.13) with various RAG performance indicators, revealing contrasts between the importance of extrinsic and intrinsic properties of passages. Semantic independence between passages proves essential for system performance with a performance gain of up to 56.2% in factual correctness and 21.1% in answer correctness. On the contrary, traditional assumptions about maintaining concept unity within passages show minimal impact. These findings provide actionable insights for optimizing chunking strategies, thus improving RAG system design to produce more factually correct responses.

cs.CL↗

arXiv2025

AutoMix: Automatically Mixing Language Models

Large language models (LLMs) are now available from cloud API providers in various sizes and configurations. While this diversity offers a broad spectrum of choices, effectively leveraging the options to optimize computational cost and performance remains challenging. In this work, we present Automix, an approach that strategically routes queries to larger LMs, based on the approximate correctness of outputs from a smaller LM. Central to Automix are two key technical contributions. First, it has a few-shot self-verification mechanism, which estimates the reliability of its own outputs without requiring extensive training. Second, given that self-verification can be noisy, it employs a POMDP based router that can effectively select an appropriately sized model, based on answer confidence. Experiments across five language models and five challenging datasets show that Automix consistently surpasses strong baselines, reducing computational cost by over 50% for comparable performance.

cs.CL↗

arXiv2024

On a probabilistic global optimizer derived from the Walker slice sampling

This article presents a zeroth order probabilistic global optimization algorithm -- SwiftNav -- for (not necessarily convex) functions over a compact domain. A discretization procedure is deployed on the compact domain, starting with a small step-size $h > 0$ and subsequently adaptively refining it in the course of a simulated annealing routine utilizing the Walker slice and the Gibbs sampler, in order to identify a set of global optimizers up to good precision. SwiftNav is parallelizable, which helps with scalability as the dimension of decision variables increases. Several numerical experiments are included here to demonstrate the effectiveness and accuracy of SwiftNav in high-dimensional benchmark optimization problems.

math.OC↗