arXiv subjects

Harsh Sharma

Publications and source records attributed to Harsh Sharma.

At least 19 recordsLinked to original sources

arXiv2026

FiMI Banking: A Sovereign Model for Indian Retail Banking

Banks need conversational systems that can answer product questions, assist customers with account-related requests, and operate safely within strict operational and regulatory constraints. General-purpose language models do not reliably meet these requirements. They fall short when a task requires grounded information, correct tool use, or cautious handling of bank-specific sensitive situations. We introduce FiMI Banking, a controlled Indian retail-banking setting. We build it from vetted banking documents, structured ground truth, synthetic customer backgrounds, and banking tools. We evaluate two post-training approaches: preference optimization for response-level behavior, and reinforcement learning with verifiable rewards for multi-turn tool-use tasks. Preference optimization improves safe behavior substantially: out-of-scope refusal rises from 52% to 80%. Reinforcement learning improves edge-case performance from 0.509 to 0.718 and order-sensitive task performance from 0.590 to 0.679, while using 29% fewer generated tokens. These results show that preference optimization and verifiable-reward reinforcement learning address complementary requirements for reliable banking agents.

cs.AI↗

arXiv2026

Jury Duty: Calibration and Orientation Failures in MLLM-as-a-Judge Under Cultural Ambiguity

MLLM-as-a-Judge is conventionally validated by agreement with human annotations, but this metric is undefined when the human pool is culturally heterogeneous. We introduce VOIR DIRE, a multimodal benchmark of 626 culturally paired image--prompt artifacts spanning U.S. and mainland Chinese contexts across food, fashion, and architecture, with annotator pools that are within-pool reliable (a = 0.86/0.74) but cross-pool divergent on evaluation (Q1 r = -0.12). Across six MLLMs, the bias decomposes into two failures: a positivity-floor calibration failure (compressed scale use) and an orientation failure (default to one cultural norm). On this corpus, where contested items are sampled to split the two pools, the floor mechanically validates the more-permissive Chinese reading; persona prompting partially recovers calibration, but the orientation residual survives, evidence the tilt is not reducible to scale compression. Reference-pool in-context demonstrations deepen the orientation residual and inflate the high end rather than restoring use of the low end. Model origin adds a small additive tilt (~0.10 MAE) that is approximately invariant under demonstration. We recommend reporting alignment against each reference pool separately and treating cross-pool divergence as a judge property.

cs.CV↗

arXiv2026

BRIDGE and TCH-Net: Heterogeneous Benchmark and Multi-Branch Baseline for Cross-Domain IoT Botnet Detection

IoT botnet detection has advanced, yet most published systems are validated on a single dataset and rarely generalise across environments. Heterogeneous feature spaces make multi-dataset training practically impossible without discarding semantic interpretability or introducing data integrity violations. No prior work has addressed both problems with a formally specified, reproducible methodology. This paper does. We introduce BRIDGE (Benchmark Reference for IoT Domain Generalisation Evaluation), the first formally specified heterogeneous multi-dataset benchmark for IoT intrusion detection, unifying CICIDS-2017, CIC-IoT-2023, Bot-IoT, Edge-IIoTset, and N-BaIoT through a 46-feature semantic canonical vocabulary grounded in CICFlowMeter nomenclature, with genuine-equivalence-only feature mapping, explicit zero-filling, and per-dataset coverage from 15% to 93%. A leave-one-dataset-out (LODO) protocol makes the generalisation gap precisely measurable: all five evaluated architectures achieve mean LODO F1 between 0.39 and 0.47, and we establish the first community generalisation baseline at mean LODO F1 = 0.5577, a result that shifts the agenda from single-benchmark optimisation toward cross-environment generalisation. We propose TCH-Net, a multi-branch network fusing a three-path Temporal branch (residual convolutional-BiGRU, stride-downsampled BiGRU, pre-LayerNorm Transformer), a provenance-conditioned Contextual branch, and a Statistical branch via Cross-Branch Gated Attention Fusion (CB-GAF) with learnable sigmoid gates for dynamic feature-wise mixing. Across five random seeds, TCH-Net achieves F1 = 0.8296 +/- 0.0028, AUC = 0.9380 +/- 0.0025, and MCC = 0.6972 +/- 0.0056, outperforming all twelve baselines (p < 0.05, Wilcoxon) and recording the highest LODO F1 overall. BRIDGE and the full pipeline are at https://github.com/Ammar-ss/TCH-Net.

cs.CR↗

arXiv2026

Parameter-Efficient Token Embedding Editing for Clinical Class-Level Unlearning

Machine unlearning is increasingly important for clinical language models, where privacy regulations and institutional policies may require removing sensitive information from deployed systems without retraining from scratch. In practice, deletion requests must balance effective forgetting of targeted information with preservation of model utility and minimal parameter modification. We introduce Sparse Token Embedding Unlearning (STEU), a parameter-efficient method for behavioral class-level unlearning that updates only PMI-selected token embeddings together with a small classifier head while keeping all encoder layers frozen. Across experiments on MIMIC-IV, MIMIC-III, and eICU using BioClinicalBERT, BERT-base, and DistilBERT, STEU consistently suppresses the target class while largely preserving retained task performance. In the primary MIMIC-IV setting, STEU achieves near-complete forgetting (forget F1 = 0.0004) while maintaining competitive retained utility (retain avg F1 = 0.4766) after modifying only 0.19\% of model parameters. These results suggest that targeted behavioral unlearning can be achieved through sparse embedding edits without modifying deeper encoder representations.

cs.LG↗

arXiv2026

FiMI: A Domain-Specific Language Model for Indian Finance Ecosystem

We present FiMI (Finance Model for India), a domain-specialized financial language model developed by National Payments Corporation of India (NPCI) for Indian digital payment systems. We develop two model variants: FiMI Base and FiMI Instruct. FiMI adapts the Mistral Small 24B architecture through a multi-stage training pipeline, beginning with continuous pre-training on 68 Billion tokens of curated financial, multilingual (English, Hindi, Hinglish), and synthetic data. This is followed by instruction fine-tuning and domain-specific supervised fine-tuning focused on multi-turn, tool-driven conversations that model real-world workflows, such as transaction disputes and mandate lifecycle management. Evaluations reveal that FiMI Base achieves a 20\% improvement over the Mistral Small 24B Base model on finance reasoning benchmark, while FiMI Instruct outperforms the Mistral Small 24B Instruct model by 87\% on domain-specific tool-calling. Moreover, FiMI achieves these significant domain gains while maintaining comparable performance to models of similar size on general benchmarks.

cs.AI↗

arXiv2025

From Facts to Conclusions : Integrating Deductive Reasoning in Retrieval-Augmented LLMs

Retrieval-Augmented Generation (RAG) grounds large language models (LLMs) in external evidence, but fails when retrieved sources conflict or contain outdated or subjective information. Prior work address these issues independently but lack unified reasoning supervision. We propose a reasoning-trace-augmented RAG framework that adds structured, interpretable reasoning across three stages : (1) document-level adjudication, (2) conflict analysis, and (3) grounded synthesis, producing citation-linked answers or justified refusals. A Conflict-Aware Trust-Score (CATS) pipeline is introduced which evaluates groundedness, factual correctness, refusal accuracy, and conflict-behavior alignment using an LLM-as-a-Judge. Our 539-query reasoning dataset and evaluation pipeline establish a foundation for conflict-aware, interpretable RAG systems. Experimental results demonstrate substantial gains over baselines, most notably with Qwen, where Supervised Fine-Tuning improved End-to-End answer correctness from 0.069 to 0.883 and behavioral adherence from 0.074 to 0.722.

cs.CL↗

arXiv2025

Quadratic power enhancement in extended Dicke quantum battery

We demonstrate a quadratic enhancement of power in a battery consisting of $N$ two-level systems or spins interacting with two photonic cavity modes, where one of the modes is in the dispersive regime. In contrast to Dicke batteries, the power enhancement arises from a $N^2$ scaling of both quantum correlations and speed of evolution, thus highlighting genuine quantum advantage. Moreover, this hybrid setup is experimentally realizable and ensures that power enhancement is not achieved at significant cost to energy efficiency, while allowing for greater tunability and stable operation in the presence of noise.

quant-ph↗

arXiv2025

Dynamic Shape Control of Soft Robots Enabled by Data-Driven Model Reduction

Soft robots have shown immense promise in settings where they can leverage dynamic control of their entire bodies. However, effective dynamic shape control requires a controller that accounts for the robot's high-dimensional dynamics--a challenge exacerbated by a lack of general-purpose tools for modeling soft robots amenably for control. In this work, we conduct a comparative study of data-driven model reduction techniques for generating linear models amendable to dynamic shape control. We focus on three methods--the eigensystem realization algorithm, dynamic mode decomposition with control, and the Lagrangian operator inference (LOpInf) method. Using each class of model, we explored their efficacy in model predictive control policies for the dynamic shape control of a simulated eel-inspired soft robot in three experiments: 1) tracking simulated reference trajectories guaranteed to be feasible, 2) tracking reference trajectories generated from a biological model of eel kinematics, and 3) tracking reference trajectories generated by a reduced-scale physical analog. In all experiments, the LOpInf-based policies generated lower tracking errors than policies based on other models.

cs.RO↗

arXiv2025

Thermodynamic Probes of Multipartite Entanglement in Strongly Interacting Quantum Systems

Quantifying multipartite entanglement in quantum many-body systems and hybrid quantum computing architectures is a fundamental yet challenging task. In recent years, thermodynamic quantities such as the maximum extractable work from an isolated system (the ergotropy) have allowed for entanglement measures that are operationally more accessible. However, these measures can be restrictive when applied to systems governed by Hamiltonians with strong collective or interparticle interactions. Motivated by advances in quantum simulators, we propose a framework that circumvents these restrictions by evaluating global and local ergotropy either through controlled quenching of interactions or by measuring suitable local observables only. We show that this formalism allows us to correctly estimate genuine multipartite entanglement in both stationary and time-evolved states of systems with strong interactions, including parametrized quantum states simulated on a quantum circuit with varying circuit depth and noise. We demonstrate its applicability to realistic physical models, namely, the Tavis-Cummings model, the three-level Dicke model, and the transverse-field Ising model, highlighting its potential as a versatile tool for characterizing entanglement in near-term quantum simulators.

quant-ph↗

arXiv2025

CHIPSIM: A Co-Simulation Framework for Deep Learning on Chiplet-Based Systems

Due to reduced manufacturing yields, traditional monolithic chips cannot keep up with the compute, memory, and communication demands of data-intensive applications, such as rapidly growing deep neural network (DNN) models. Chiplet-based architectures offer a cost-effective and scalable solution by integrating smaller chiplets via a network-on-interposer (NoI). Fast and accurate simulation approaches are critical to unlocking this potential, but existing methods lack the required accuracy, speed, and flexibility. To address this need, this work presents CHIPSIM, a comprehensive co-simulation framework designed for parallel DNN execution on chiplet-based systems. CHIPSIM concurrently models computation and communication, accurately capturing network contention and pipelining effects that conventional simulators overlook. Furthermore, it profiles the chiplet and NoI power consumptions at microsecond granularity for precise transient thermal analysis. Extensive evaluations with homogeneous/heterogeneous chiplets and different NoI architectures demonstrate the framework's versatility, up to 340% accuracy improvement, and power/thermal analysis capability.

cs.AR↗

arXiv2025

Taming the Tail: NoI Topology Synthesis for Mixed DL Workloads on Chiplet-Based Accelerators

Heterogeneous chiplet-based systems improve scaling by disag-gregating CPUs/GPUs and emerging technologies (HBM/DRAM).However this on-package disaggregation introduces a latency inNetwork-on-Interposer(NoI). We observe that in modern large-modelinference, parameters and activations routinely move backand forth from HBM/DRAM, injecting large, bursty flows into theinterposer. These memory-driven transfers inflate tail latency andviolate Service Level Agreements (SLAs) across k-ary n-cube base-line NoI topologies. To address this gap we introduce an InterferenceScore (IS) that quantifies worst-case slowdown under contention.We then formulate NoI synthesis as a multi-objective optimization(MOO) problem. We develop PARL (Partition-Aware ReinforcementLearner), a topology generator that balances throughput, latency,and power. PARL-generated topologies reduce contention at the memory cut, meet SLAs, and cut worst-case slowdown to 1.2 times while maintaining competitive mean throughput relative to link-rich meshes. Overall, this reframes NoI design for heterogeneouschiplet accelerators with workload-aware objectives.

cs.AR↗

arXiv2025

NVIDIA Nemotron Nano 2: An Accurate and Efficient Hybrid Mamba-Transformer Reasoning Model

We introduce Nemotron-Nano-9B-v2, a hybrid Mamba-Transformer language model designed to increase throughput for reasoning workloads while achieving state-of-the-art accuracy compared to similarly-sized models. Nemotron-Nano-9B-v2 builds on the Nemotron-H architecture, in which the majority of the self-attention layers in the common Transformer architecture are replaced with Mamba-2 layers, to achieve improved inference speed when generating the long thinking traces needed for reasoning. We create Nemotron-Nano-9B-v2 by first pre-training a 12-billion-parameter model (Nemotron-Nano-12B-v2-Base) on 20 trillion tokens using an FP8 training recipe. After aligning Nemotron-Nano-12B-v2-Base, we employ the Minitron strategy to compress and distill the model with the goal of enabling inference on up to 128k tokens on a single NVIDIA A10G GPU (22GiB of memory, bfloat16 precision). Compared to existing similarly-sized models (e.g., Qwen3-8B), we show that Nemotron-Nano-9B-v2 achieves on-par or better accuracy on reasoning benchmarks while achieving up to 6x higher inference throughput in reasoning settings like 8k input and 16k output tokens. We are releasing Nemotron-Nano-9B-v2, Nemotron-Nano12B-v2-Base, and Nemotron-Nano-9B-v2-Base checkpoints along with the majority of our pre- and post-training datasets on Hugging Face.

cs.CL↗

arXiv2025

Group Fair Matchings using Convex Cost Functions

We consider the problem of assigning items to platforms where each item has a utility associated with each of the platforms to which it can be assigned. Each platform has a soft constraint over the total number of items it serves, modeled via a convex cost function. Additionally, items are partitioned into groups, and each platform also incurs group-specific convex cost over the number of items from each group that can be assigned to the platform. These costs promote group fairness by penalizing imbalances, yielding a soft variation of fairness notions introduced in prior work, such as Restricted Dominance and Minority protection. Restricted Dominance enforces upper bounds on group representation, while Minority protection enforces lower bounds. Our approach replaces such hard constraints with cost-based penalties, allowing more flexible trade-offs. Our model also captures Nash Social Welfare kind of objective. The cost of an assignment is the sum of the values of all the cost functions across all the groups and platforms. The objective is to find an assignment that minimizes the cost while achieving a total utility that is at least a user-specified threshold. The main challenge lies in balancing the overall platform cost with group-specific costs, both governed by convex functions, while meeting the utility constraint. We present an efficient polynomial-time approximation algorithm, supported by theoretical guarantees and experimental evaluation. Our algorithm is based on techniques involving linear programming and network flows. We also provide an exact algorithm for a special case with uniform utilities and establish the hardness of the general problem when the groups can intersect arbitrarily.

cs.GT↗

arXiv2025

THERMOS: Thermally-Aware Multi-Objective Scheduling of AI Workloads on Heterogeneous Multi-Chiplet PIM Architectures

Chiplet-based integration enables large-scale systems that combine diverse technologies, enabling higher yield, lower costs, and scalability, making them well-suited to AI workloads. Processing-in-Memory (PIM) has emerged as a promising solution for AI inference, leveraging technologies such as ReRAM, SRAM, and FeFET, each offering unique advantages and trade-offs. A heterogeneous chiplet-based PIM architecture can harness the complementary strengths of these technologies to enable higher performance and energy efficiency. However, scheduling AI workloads across such a heterogeneous system is challenging due to competing performance objectives, dynamic workload characteristics, and power and thermal constraints. To address this need, we propose THERMOS, a thermally-aware, multi-objective scheduling framework for AI workloads on heterogeneous multi-chiplet PIM architectures. THERMOS trains a single multi-objective reinforcement learning (MORL) policy that is capable of achieving Pareto-optimal execution time, energy, or a balanced objective at runtime, depending on the target preferences. Comprehensive evaluations show that THERMOS achieves up to 89% faster average execution time and 57% lower average energy consumption than baseline AI workload scheduling algorithms with only 0.14% runtime and 0.022% energy overhead.

cs.AR↗

arXiv2025

ThematicPlane: Bridging Tacit User Intent and Latent Spaces for Image Generation

Generative AI has made image creation more accessible, yet aligning outputs with nuanced creative intent remains challenging, particularly for non-experts. Existing tools often require users to externalize ideas through prompts or references, limiting fluid exploration. We introduce ThematicPlane, a system that enables users to navigate and manipulate high-level semantic concepts (e.g., mood, style, or narrative tone) within an interactive thematic design plane. This interface bridges the gap between tacit creative intent and system control. In our exploratory study (N=6), participants engaged in divergent and convergent creative modes, often embracing unexpected results as inspiration or iteration cues. While they grounded their exploration in familiar themes, differing expectations of how themes mapped to outputs revealed a need for more explainable controls. Overall, ThematicPlane fosters expressive, iterative workflows and highlights new directions for intuitive, semantics-driven interaction in generative design tools.

cs.HC↗

arXiv2025

Designing High-Performance and Thermally Feasible Multi-Chiplet Architectures enabled by Non-bendable Glass Interposer

Multi-chiplet architectures enabled by glass interposer offer superior electrical performance, enable higher bus widths due to reduced crosstalk, and have lower capacitance in the redistribution layer than current silicon interposer-based systems. These advantages result in lower energy per bit, higher communication frequencies, and extended interconnect range. However, deformation of the package (warpage) in glass interposer-based systems becomes a critical challenge as system size increases, leading to severe mechanical stress and reliability concerns. Beyond a certain size, conventional packaging techniques fail to manage warpage effectively, necessitating new approaches to mitigate warpage induced bending with scalable performance for glass interposer based multi-chiplet systems. To address these inter-twined challenges, we propose a thermal-, warpage-, and performance-aware design framework that employs architecture and packaging co-optimization. The proposed framework disintegrates the surface and embedded chiplets to balance conflicting design objectives, ensuring optimal trade-offs between performance, power, and structural reliability. Our experiments demonstrate that optimized multi-chiplet architectures from our design framework achieve up to 64.7% performance improvement and 40% power reduction compared to traditional 2.5D systems to execute deep neural network workloads with lower fabrication costs.

cs.AR↗

arXiv2025

Beyond Sensor Data: Foundation Models of Behavioral Data from Wearables Improve Health Predictions

Wearable devices record physiological and behavioral signals that can improve health predictions. While foundation models are increasingly used for such predictions, they have been primarily applied to low-level sensor data, despite behavioral data often being more informative due to their alignment with physiologically relevant timescales and quantities. We develop foundation models of such behavioral signals using over 2.5B hours of wearable data from 162K individuals, systematically optimizing architectures and tokenization strategies for this unique dataset. Evaluated on 57 health-related tasks, our model shows strong performance across diverse real-world applications including individual-level classification and time-varying health state prediction. The model excels in behavior-driven tasks like sleep prediction, and improves further when combined with representations of raw sensor data. These results underscore the importance of tailoring foundation model design to wearables and demonstrate the potential to enable new health applications.

cs.LG↗

arXiv2025

Structure-preserving Lift & Learn: Scientific machine learning for nonlinear conservative partial differential equations

This work presents structure-preserving Lift & Learn, a scientific machine learning method that employs lifting variable transformations to learn structure-preserving reduced-order models for nonlinear partial differential equations (PDEs) with conservation laws. We propose a hybrid learning approach based on a recently developed energy-quadratization strategy that uses knowledge of the nonlinearity at the PDE level to derive an equivalent quadratic lifted system with quadratic system energy. The lifted dynamics obtained via energy quadratization are linear in the old variables, making model learning very effective in the lifted setting. Based on the lifted quadratic PDE model form, the proposed method derives quadratic reduced terms analytically and then uses those derived terms to formulate a constrained optimization problem to learn the remaining linear reduced operators in a structure-preserving way. The proposed hybrid learning approach yields computationally efficient quadratic reduced-order models that respect the underlying physics of the high-dimensional problem. We demonstrate the generalizability of quadratic models learned via the proposed structure-preserving Lift & Learn method through three numerical examples: the one-dimensional wave equation with exponential nonlinearity, the two-dimensional sine-Gordon equation, and the two-dimensional Klein-Gordon-Zakharov equations. The numerical results show that the proposed learning approach is competitive with the state-of-the-art structure-preserving data-driven model reduction method in terms of both accuracy and computational efficiency.

cs.LG↗