arXiv subjects

Abhishek Dey

Publications and source records attributed to Abhishek Dey.

16 recordsLinked to original sources

arXiv2026

ComplianceGate: Classifier-Gated Multi-Tier LLM Routing for Inference in Regulated Industries

Large language models deployed in regulated industries operate under two constraints: compliance enforcement and cost efficiency. Personally identifiable information (PII) in user queries can reach model endpoints before the system determines whether that data should leave its jurisdictional boundary. Serving all queries through a single large model consumes full GPU capacity regardless of query complexity while offering no mechanism for geographic routing. Mixture-of-Experts architectures do not address this routing occurs between expert layers within the model after data has already arrived at the endpoint, with all experts loaded in memory regardless of query complexity. We propose a classifier-gated routing architecture that enforces compliance by design. A trained encoder classifier sits before any decoder inference, evaluating each query for complexity and data sensitivity, then routing it to an appropriately sized dense model in the appropriate geographic location. PII-containing queries route to local endpoints before any LLM computation begins, making data residency violations structurally impossible. Simple queries reach small, fast models at a fraction of the cost. Our evaluation on 600 queries demonstrates 39% median latency reduction, 33-52% cost savings depending on query distribution, and generation throughput of 122-200 tokens/second versus 50-64 for the baseline. The encoder classifier achieves 99.2% accuracy with near-perfect PII recall at 7ms inference overhead, establishing pre-inference classification as a practical path to compliance-by-design LLM deployment.

cs.LG↗

arXiv2026

An Exponentially stable Extended Kalman Filter with Estimate dependent Process noise Covariance for Chemical Reaction Networks

Biomolecular systems are often modeled with partially known nonlinear stochastic dynamics, making state and parameter estimation a central challenge. While Kalman filtering techniques are widely used in this setting, their performance critically depends on the choice of the process noise covariance, which is typically assumed constant and heuristically tuned. Such assumptions are not justified for biomolecular systems, where intrinsic noise arises from underlying reaction kinetics. In previous works, a process noise covariance update based on the Chemical Langevin Equation (CLE) was introduced for Extended Kalman Filter (EKF)-based estimation in Chemical Reaction Networks (CRN). In this work, we analyze the stochastic stability of this filtering framework. In particular, we obtain a conservative upper bound on sampling interval for discrete-time biomolecular systems that ensures mean-square exponential boundedness under stated assumptions. The proposed framework is validated through simulations on a nonlinear gene expression model. The analysis provides theoretical justification for CLE-based process noise covariance modeling in EKF design for biomolecular circuits, reducing reliance on heuristic covariance tuning.

eess.SY↗

arXiv2026

Batch Prompting Suppresses Overthinking Reasoning Under Constraint: How Batch Prompting Suppresses Overthinking in Reasoning Models

Large Reasoning Models (LRMs) achieve strong performance through explicit chain-of-thought reasoning but suffer from \textit{overthinking}: generating excessive reasoning tokens even for trivial queries. {Beyond inflating cost, overthinking can be self-defeating: models enter recursive self-doubt loops that exhaust token budgets without producing an answer, causing API timeouts that directly hurt accuracy.} We present an empirical study showing that \textbf{batch prompting}, originally introduced for throughput optimization, effectively suppresses overthinking at inference time. Across 13 diverse benchmarks with DeepSeek-R1 and OpenAI-o1, batch prompting {reduces reasoning tokens by 76\% (2{,}950$\mapsto$710), on average, while preserving or improving accuracy}. Through behavioral analysis, we find that batching induces three beneficial effects: (1) it reduces per-query reasoning effort when multiple queries share a context; (2) it enables pattern induction, where models generalize from earlier examples to solve later ones; and (3) it suppresses hedging behavior (e.g., ``\texttt{wait,}'' ``\texttt{let me double-check}'') that signals metacognitive loops. We also show that explicit prompt constraints (``\texttt{Use no more than 100 tokens in thinking.}'') fail to reduce overthinking; models either ignore them or sacrifice accuracy. These findings reframe batch prompting as more than a cost optimization: it is a practical inference-time technique that improves efficiency and reliability without model modification.

cs.CL↗

arXiv2025

Real-Time Performance Benchmarking of TinyML Models in Embedded Systems (PICO: Performance of Inference, CPU, and Operations)

This paper presents PICO-TINYML-BENCHMARK, a modular and platform-agnostic framework for benchmarking the real-time performance of TinyML models on resource-constrained embedded systems. Evaluating key metrics such as inference latency, CPU utilization, memory efficiency, and prediction stability, the framework provides insights into computational trade-offs and platform-specific optimizations. We benchmark three representative TinyML models -- Gesture Classification, Keyword Spotting, and MobileNet V2 -- on two widely adopted platforms, BeagleBone AI64 and Raspberry Pi 4, using real-world datasets. Results reveal critical trade-offs: the BeagleBone AI64 demonstrates consistent inference latency for AI-specific tasks, while the Raspberry Pi 4 excels in resource efficiency and cost-effectiveness. These findings offer actionable guidance for optimizing TinyML deployments, bridging the gap between theoretical advancements and practical applications in embedded systems.

cs.SE↗

arXiv2025

Multimodal Event Detection: Current Approaches and Defining the New Playground through LLMs and VLMs

In this paper, we study the challenges of detecting events on social media, where traditional unimodal systems struggle due to the rapid and multimodal nature of data dissemination. We employ a range of models, including unimodal ModernBERT and ConvNeXt-V2, multimodal fusion techniques, and advanced generative models like GPT-4o, and LLaVA. Additionally, we also study the effect of providing multimodal generative models (such as GPT-4o) with a single modality to assess their efficacy. Our results indicate that while multimodal approaches notably outperform unimodal counterparts, generative approaches despite having a large number of parameters, lag behind supervised methods in precision. Furthermore, we also found that they lag behind instruction-tuned models because of their inability to generate event classes correctly. During our error analysis, we discovered that common social media issues such as leet speak, text elongation, etc. are effectively handled by generative approaches but are hard to tackle using supervised approaches.

cs.CL↗

arXiv2024

ResEmoteNet: Bridging Accuracy and Loss Reduction in Facial Emotion Recognition

The human face is a silent communicator, expressing emotions and thoughts through its facial expressions. With the advancements in computer vision in recent years, facial emotion recognition technology has made significant strides, enabling machines to decode the intricacies of facial cues. In this work, we propose ResEmoteNet, a novel deep learning architecture for facial emotion recognition designed with the combination of Convolutional, Squeeze-Excitation (SE) and Residual Networks. The inclusion of SE block selectively focuses on the important features of the human face, enhances the feature representation and suppresses the less relevant ones. This helps in reducing the loss and enhancing the overall model performance. We also integrate the SE block with three residual blocks that help in learning more complex representation of the data through deeper layers. We evaluated ResEmoteNet on four open-source databases: FER2013, RAF-DB, AffectNet-7 and ExpW, achieving accuracies of 79.79%, 94.76%, 72.39% and 75.67% respectively. The proposed network outperforms state-of-the-art models across all four databases. The source code for ResEmoteNet is available at https://github.com/ArnabKumarRoy02/ResEmoteNet.

cs.CV↗

arXiv2024

Mutual Information Assisted Ensemble Recommender System for Identifying Critical Risk Factors in Healthcare Prognosis

Purpose: Health recommenders act as important decision support systems, aiding patients and medical professionals in taking actions that lead to patients' well-being. These systems extract the information which may be of particular relevance to the end-user, helping them in making appropriate decisions. The present study proposes a feature recommender, as a part of a disease management system, that identifies and recommends the most important risk factors for an illness. Methods: A novel mutual information and ensemble-based feature ranking approach for identifying critical risk factors in healthcare prognosis is proposed. Results: To establish the effectiveness of the proposed method, experiments have been conducted on four benchmark datasets of diverse diseases (clear cell renal cell carcinoma (ccRCC), chronic kidney disease, Indian liver patient, and cervical cancer risk factors). The performance of the proposed recommender is compared with four state-of-the-art methods using recommender systems' performance metrics like average precision@K, precision@K, recall@K, F1@K, reciprocal rank@K. The method is able to recommend all relevant critical risk factors for ccRCC. It also attains a higher accuracy (96.6% and 98.6% using support vector machine and neural network, respectively) for ccRCC staging with a reduced feature set as compared to existing methods. Moreover, the top two features recommended using the proposed method with ccRCC, viz. size of tumor and metastasis status, are medically validated from the existing TNM system. Results are also found to be superior for the other three datasets. Conclusion: The proposed recommender can identify and recommend risk factors that have the most discriminating power for detecting diseases.

cs.LG↗

arXiv2018

A Kalman Filter Approach for Biomolecular Systems with Noise Covariance Updating

An important part of system modeling is determining parameter values, particularly for biomolecular systems, where direct measurements of individual parameters are typically hard. While Extended Kalman Filters have been used for this purpose, the choice of the process noise covariance is generally unclear. In this chapter, we address this issue for biomolecular systems using a combination of Monte Carlo simulations and experimental data, exploiting the dependence of the process noise covariance on the states and parameters, as given in the Langevin framework. We adapt a Hybrid Extended Kalman Filtering technique by updating the process noise covariance at each time step based on estimates. We compare the performance of this framework with different fixed values of process noise covariance in biomolecular system models, including an oscillator model, as well as in experimentally measured data for a negative transcriptional feedback circuit. We find that the Extended Kalman Filter with such process noise covariance update is closer to the optimality condition in the sense that the innovation sequence becomes white and in achieving a balance between the mean square estimation error and parameter convergence time. The results of this chapter may help in the use of Extended Kalman Filters for systems where process noise covariance depends on states and/or parameters.

q-bio.QM↗

arXiv2018

Period-Amplitude Co-variation in Biomolecular Oscillators

The period and amplitude of biomolecular oscillators are functionally important properties in multiple contexts. For a biomolecular oscillator, the overall constraints in how tuning of amplitude affects period, and vice versa, are generally unclear. Here we investigate this co-variation of the period and amplitude in mathematical models of biomolecular oscillators using both simulations and analytical approximations. We computed the amplitude-period co-variation of eleven benchmark biomolecular oscillators as their parameters were individually varied around a nominal value, classifying the various co-variation patterns such as a simultaneous increase/ decrease in period and amplitude. Next, we repeated the classification using a power norm-based amplitude metric, to account for the amplitudes of the many biomolecular species that may be part of the oscillations, finding largely similar trends. Finally, we calculate "scaling laws" of period-amplitude co-variation for a subset of these benchmark oscillators finding that as the approximated period increases, the upper bound of the amplitude increases, or reaches a constant value. Based on these results, we discuss the effect of different parameters on the type of period-amplitude co-variation as well as the difficulty in achieving an oscillation with large amplitude and small period.

q-bio.MN↗

arXiv2017

Describing Function-based Approximations of Biomolecular Systems

Mathematical methods provide useful framework for the analysis and design of complex systems. In newer contexts such as biology, however, there is a need to both adapt existing methods as well as to develop new ones. Using a combination of analytical and computational approaches, we adapt and develop the method of describing functions to represent the input-output responses of biomolecular signalling systems. We approximate representative systems exhibiting various saturating and hysteretic dynamics in a way that is better than the standard linearization. Further, we develop analytical upper bounds for the computational error estimates. Finally, we use these error estimates to augment the limit cycle analysis with a simple and quick way to bound the predicted oscillation amplitude. These results provide system approximations that can add more insight into the local behaviour of these systems than standard linearization, compute responses to other periodic inputs, and to analyze limit cycles.

q-bio.MN↗

arXiv2017

Exploring Charge and Color Breaking vacuum in Non-Holomorphic MSSM

Non-Holomorphic MSSM (NHSSM) shows various promising features that are not easily obtained in MSSM. However, the additional Non-Holomorphic (NH) trilinear interactions that attribute to the interesting phenomenological features, also modify the effective scalar potential of the model significantly. We derive analytic constraints involving trilinear parameters $A_t'$ and $A_b'$ that exclude global charge and color breaking minima (CCB). Since the analytic constraints are obtained considering specific directions in the multi-dimensional field space, we further probe the applicability of these constraints by exhaustive scan over NH parameter space with two different regimes of $\tanβ$ and delineate the nature of metastability by considering vacuum expectation values for third generation squarks. We adhere to a natural scenario by fixing Higgsino mass parameter ($μ$) to a low value and estimate the allowed ranges of NH trilinear parameters by considering vacuum stability and observed properties of Higgs as the determining criteria.

hep-ph↗

arXiv2017

Exploring viable vacua of the $Z_3$-symmetric NMSSM

We explore the vacua of the $Z_3$-symmetric Next-to-Minimal Supersymmetric Standard Model (NMSSM) and their stability by going beyond the simplistic paradigm that works with a tree-level neutral scalar potential and adheres to some specific flat directions in the field space. Key effects are demonstrated by first studying the profiles of this potential under various circumstances of physical interest via a semi-analytical approach. The results thereof are compared to the ones obtained from a dedicated package like \veva ~which further incorporates the thermal effects to the potential. Regions of the phenomenological NMSSM (pNMSSM) parameter space that render the desired symmetry breaking (DSB) vacuum absolutely stable, long- or short-lived (in relation to the age of the Universe) under quantum/thermal tunneling are delineated. Regions that result in color and charge breaking (CCB) minima are also presented. It is demonstrated that light singlet scalars along with a light LSP (lightest supersymmetric particle) having an appreciable singlino admixture are compatible with a viable DSB vacuum and are much relevant for the collider experiments.

hep-ph↗

arXiv2016

Probing Non-holomorphic MSSM via precision constraints, dark matter and LHC data

In this analysis we explore the phenomenological constraints of models with non-holomorphic soft SUSY breaking terms in a beyond the MSSM scenario having identical particle content. The model referred as NHSSM shows various promising features like the possibility of a strong reduction in electroweak fine-tuning even for a scenario of a heavy higgsino type of LSP, a fact that is unavailable in pMSSM models. The other important aspect is satisfying the muon $g-2$ data even for a small $\tanβ$ via a small value of coupling $A_μ'$ associated with the tri-linear non-holomorphic soft term. Thus, a large SUSY contribution to muon $g-2$ is possible even for a significantly large smuon mass $m_{\tilde {μ_1}}$. The Higgs mass radiative corrections are contributed by both the holomorphic and non-holomorphic trilinear soft parameters $A_t$ and $A_t'$, thus diluting the requirement to have a larger $A_t$ to satisfy the Higgs mass data. The model also provides with valid parameter space satisfying the constraint of $B \rightarrow X_s +γ$ for large values of $\tanβ$, a scenario unfavourable in pMSSM.

hep-ph↗

arXiv2014

Exploring MSSM for Charge and Color Breaking and Other Constraints in the Context of Higgs@125 GeV

Exploring MSSM parameter space after the discovery of Higgs Boson with mass 125 GeV naturally demands large top-squark mixing or large trilinear coupling parameter $A_t$ in particular, so as to avoid excessively heavy squark, specially for the universal models like CMSSM. We study stability of electroweak symmetry breaking vacua in possible presence of deeper charge-color symmetry breaking minima within MSSM. Besides stable vacua, we consider scenarios characterized by the presence of global CCB minima, with SM like charge and color conserving vacuum, having stability over cosmologically large lifetime {(\it long-lived states)}. We allow vacuum expectation values for both stop as well as sbottom fields, since these belong to the third generation of sfermions with larger Yukawa couplings that have immediate effect on the tunneling time. Moreover, for large $μ$ regions, radiative corrections to Higgs boson mass from bottom-squark loop is quite significant. Regions of MSSM parameters space become viable for large $A_t$ and large $μ$ zones which are generically excluded via the traditional analytical CCB constraints. For a large value of $\tanβ$, safe vacua associated with large values of $|μ|$ and $|A_t|$ are predominantly long-lived and may be associated with relatively light stop masses. We also identify low $μ$ regions associated with long-lived states. Both the above zones can be friendly to muon $g-2$ constraint. We also impose constraints from ${\rm Br}(B \rightarrow X_s γ)$ and ${\rm Br}(B_s \rightarrow μ^+ μ^-)$. We do the analysis for a moderate and a large $\tanβ$. Results are compatible with dark matter related constraints, as expected.

hep-ph↗

arXiv2014

Tetrathiomolybdate Modified Au Electrodes: Convenient Tuning of the kinetics of Electron Transfer and its application in Electrocatalysis

Ammonium tetrathiomolybdate (ATM) spontaneously self assembles on Au electrodes forming a hydrophilic, air stable, pH (3-11) tolerant multilayer that is stable over a reasonably large potential window. The ATM functionalized Au electrodes can adsorb iron porphyrin catalysts and act as O2 reducing electrodes. These electrodes are stable enough to perform rotating disk electrochemistry (RDE) as well as rotating ring disk electrochemistry (RRDE) experiments. The X-ray photoelectron spectroscopy (XPS) data indicate that the sulphide atoms of ATM anchors a single ATM layer on to Au and the subsequent layers grow vertically due to the presence of hydrogen bonding NH4+ counter-ions. The formation and growth of these ATM adlayers is investigated using atomic force microscopy (AFM), scanning electron microscopy (SEM) and a series of electrochemical data. The ATM functionalized Au electrodes have double layer capacitance comparable to those reported for Au electrode bearing short chain alkyl thiol self assembled monolayer (SAM). Importantly, the rate of interfacial charge transfer (CT) can be tuned by controlling the thickness of the adlayers by simply adjusting the deposition time. Importantly the kinetics of a catalyst adsorbed on this ATM adlayer can be switched from mass transfer limited to CT limited regime by adjusting the deposition time.

cond-mat.mtrl-sci↗

arXiv2014

An Acetate Bound Cobalt Oxide Catalyst for Water Oxidation: Role of Monovalent Anions and Cations In Lowering Overpotential

Co(II) dissolved in acetate buffer at pH 7 is found to be a good water oxidation catalyst (WOC) showing electrocatalytic water oxidation current significantly greater than Co(II) in phosphate buffer under the same conditions owing to the higher solubility of the former. When electrodeposited on ITO/FTO electrodes it forms acetate bound cobalt(II)oxide based material (Co-Ac-WOC) showing catalytic water oxidation current density of 0.1 mA/cm$^{2}$ at 830 mV and 1 mA/cm$^{2}$ at 1 V in a pH 7 buffer solution. The morphology of Co-Ac-OEC is investigated with AFM, HR-TEM and SEM (at different times and electrodeposition potentials). The chemical composition of Co-Ac-OEC is investigated using XPS, EDX, combustion analysis and ATR-FTIR which indicates that this material has a CoO core with chloride and acetate anions bound to the Co center. Sodium is found to be integrated in the Co-Ac-WOC. The presence of the sodium ions and the chloride ions lowers the onset potential for oxygen evolution reaction (OER) by 240 mV relative to the classic Co-Pi at pH 7. The lower onset potential and higher OER current lowers the exchange current density to 10$^{-6.7}$ in Co-Ac-WOC relative to 10$^{-8}$-10$^{-10}$ in Co-Pi and its derivatives.

cond-mat.mtrl-sci↗