arXiv subjects

Subhankar Roy

Publications and source records attributed to Subhankar Roy.

At least 19 recordsLinked to original sources

arXiv2026

DistMoE: Private-data Rehearsal-free Routing in Mixture-of-Experts for Distributed Instruction Tuning

Multimodal Large Language Models (MLLMs) have shown strong multimodal instruction-following ability, but adapting them to diverse visual-language domains typically assumes centralized data access and costly joint training. This is restrictive when data is distributed across private, domain-specific, or permission-limited clients. To this end, we propose DistMoE, a mixture-of-experts (MoE) approach for distributed visual instruction tuning. In each layer of the language decoder it augments the public feedforward network (FFN) with a client-specific private FFN expert, with the goal to acquire domain-specific knowledge. However, independent expert training causes the private FFNs to learn representation of different scale and magnitudes, making merging the experts difficult. To reduce client-specific drift, we introduce a public-anchored expert composition stage that updates only routers and lightweight private projection adapters on a mix of local client data and public data, via an isotropic regularization loss, therefore making it cross-client rehearsal-free composition. During inference, DistMoE performs modular routing over public and private experts, enabling token-wise domain composition without explicit domain labels. Experiments across diverse visual-language benchmarks show that DistMoE enables flexible expert reuse, effective domain adaptation, and competitive performance while preserving modular control over client-specific knowledge. Codes are available at https://github.com/mainaksingha01/DistMoE.

cs.CV↗

arXiv2026

The Triadic Texture: Neutrino Predictions, Viable Vacuum, and Phenomenological Constraints

A minimal and predictive neutrino mass matrix texture for Majorana neutrino is proposed. The texture favours the normal hierarchy of neutrino mass eigenvalues. It further predicts the octant of $\theta_{23}$, constraints $\delta$, gives bounds on neutrino mass eigenvalues and also gives ranges for the two Majorana phases.The texture is realised in the framework of a Type-I seesaw and a Weinberg like dimension 6 operator under an extended symmetry of $SU(2)_L \otimes U(1)_Y \otimes A_4 \otimes Z_{10} \otimes Z_7 \otimes Z_5 \otimes Z_3$. The texture can be realised with different sets of vacuum expectation values of the associated scalar fields, but not all such sets lead to a viable scalar sector. The model also constrains the allowed channels of charged lepton flavour violation and leads to a suppressed baryon asymmetry generation through conventional leptogenesis.

hep-ph↗

arXiv2026

Predict-then-Diffuse: Adaptive Response Length for Compute-Budgeted Inference in Diffusion LLMs

Diffusion-based Large Language Models (D-LLMs) represent a promising frontier in generative AI, offering fully parallel token generation that can lead to significant throughput advantages and superior GPU utilization over the traditional autoregressive paradigm. However, this parallelism is constrained by the requirement of a fixed-size response length prior to generation. This architectural limitation imposes a severe trade-off: oversized response length results in computational waste on semantically meaningless padding tokens, while undersized response length causes output truncation requiring costly re-computations that introduce unpredictable latency spikes. To tackle this issue, we propose Predict-then-Diffuse, a simple and model-agnostic framework that enables compute-budgeted inference per input query by first estimating the response length and then using it to run inference with D-LLM. At its core lies an Adaptive Response Length Predictor (AdaRLP), which estimates the optimal response length given an input query. As a measure against under-estimating the response length and re-running inference with a higher value, we introduce a data-driven safety mechanism based on a small increase of the predicted length. As a whole, our framework avoids wasting computation on padding tokens, at the same time preserving output quality. Experimental validation on multiple datasets demonstrates that Predict-then-Diffuse significantly reduces computational costs (FLOP) compared to the default D-LLM inference mechanism, while being robust to skewed data distributions.

cs.LG↗

arXiv2026

ProactiveBench: Benchmarking Proactiveness in Multimodal Large Language Models

Effective collaboration begins with knowing when to ask for help. For example, when trying to identify an occluded object, a human would ask someone to remove the obstruction. Can MLLMs exhibit a similar "proactive" behavior by requesting simple user interventions? To investigate this, we introduce ProactiveBench, a benchmark built from seven repurposed datasets that tests proactiveness across different tasks such as recognizing occluded objects, enhancing image quality, and interpreting coarse sketches. We evaluate 22 MLLMs on ProactiveBench, showing that (i) they generally lack proactiveness; (ii) proactiveness does not correlate with model capacity; (iii) "hinting" at proactiveness yields only marginal gains. Surprisingly, we found that conversation histories and in-context learning introduce negative biases, hindering performance. Finally, we explore a simple fine-tuning strategy based on reinforcement learning: its results suggest that proactiveness can be learned, even generalizing to unseen scenarios. We publicly release ProactiveBench as a first step toward building proactive multimodal models.

cs.CV↗

arXiv2026

A Common Origin for Diverse Neutrino Mass Matrix Textures

We plan to decipher the common origin of neutrino mass textures within a $\Delta(27)$ based framework, in association with Type-I+II seesaw mechanism. Motivated by the diversity of texture structures used in phenomenological analyses, we examine how a single framework can reproduce them without additional assumptions. We show that the interplay between vacuum alignments, model parameters, and specific Yukawa choices naturally yields several well established neutrino mass textures in the literature. Our goal is to demonstrate that a minimal $\Delta(27)$ framework can unify these viable textures within a common theoretical origin.

hep-ph↗

arXiv2026

Superresolution technique beyond the diffraction limit under a structured beam via different optical nanostructures

To overcome the limit of diffraction while achieving the superresolution technique, solid immersion lenses are the key optical elements for data storage and nanophotonics applications. Recent demonstrations have shown how different nanostructures (such as elliptical solid immersion lenses) are used in diverse fields of increasing resolution in the presence of a structured Gaussian beam. By applying twisted beams such as angular momentum beams (Laguerre- Gaussian) and spatial higher-order Gaussian beams (Hermite- Gauss), we can attain a sharp near-field focal spot pattern, which is considerably better than the conventional solid immersion lens structure in ~mm scale specifically for imaging beyond diffraction limit. Our computation results present a resolution of ~27 nm under a specific Hermite -Gauss mode illumination on a pyramidal shape nanolens structure. By numerical simulations, tolerance has been confirmed with a slight variation in beam size and geometrical modification to make the model compatible with fabrication errors. This narrow bandwidth intensity distribution can be utilized for scanning the sample with higher resolution, especially in the field of quantum technology.

physics.optics↗

arXiv2025

How (Mis)calibrated is your Federated CLIP and what to do about it?

Vision-language models (VLMs) such as CLIP are increasingly adapted across decentralized data silos, yet the reliability of their predictions under federated learning (FL) remains largely unexplored. In this work, we present a systematic study of calibration in federated CLIP under non-IID client distributions. Our experiments reveal that widely used prompt-tuning methods consistently degrade calibration, often yielding substantially higher calibration error despite competitive recognition performance, while explicit training-time calibration regularizers provide only limited improvements. Motivated by these findings, we identify the choice of fine-tuning parameterization as a critical factor governing calibration and conduct a controlled comparison between prompt tuning and five backbone fine-tuning (BFT) strategies: AdaptFormer, LayerNorm, LoRA, VeRA, and DoRA. We find that BFT methods generally offer a more favorable accuracy-calibration trade-off than prompt tuning, although their benefits are not universal. Through extensive analysis, we show that calibration behavior is closely linked to the geometry of federated updates, residual parameterization, and the resulting client and logit drift. Across in-distribution, domain-generalization, and base-to-new evaluation settings, our results establish fine-tuning parameterization as a central design choice for building accurate and reliable federated CLIP models. Codes are available at https://github.com/mainaksingha01/FL2oRA.

cs.CV↗

arXiv2025

LT-Soups: Bridging Head and Tail Classes via Subsampled Model Soups

Real-world datasets typically exhibit long-tailed (LT) distributions, where a few head classes dominate and many tail classes are severely underrepresented. While recent work shows that parameter-efficient fine-tuning (PEFT) methods like LoRA and AdaptFormer preserve tail-class performance on foundation models such as CLIP, we find that they do so at the cost of head-class accuracy. We identify the head-tail ratio, the proportion of head to tail classes, as a crucial but overlooked factor influencing this trade-off. Through controlled experiments on CIFAR100 with varying imbalance ratio ($\rho$) and head-tail ratio ($\eta$), we show that PEFT excels in tail-heavy scenarios but degrades in more balanced and head-heavy distributions. To overcome these limitations, we propose LT-Soups, a two-stage model soups framework designed to generalize across diverse LT regimes. In the first stage, LT-Soups averages models fine-tuned on balanced subsets to reduce head-class bias; in the second, it fine-tunes only the classifier on the full dataset to restore head-class accuracy. Experiments across six benchmark datasets show that LT-Soups achieves superior trade-offs compared to both PEFT and traditional model soups across a wide range of imbalance regimes.

cs.LG↗

arXiv2025

Ensembling Pruned Attention Heads For Uncertainty-Aware Efficient Transformers

Uncertainty quantification (UQ) is essential for deploying deep neural networks in safety-critical settings. Although methods like Deep Ensembles achieve strong UQ performance, their high computational and memory costs hinder scalability to large models. We introduce Hydra Ensembles, an efficient transformer-based ensemble that prunes attention heads to create diverse members and merges them via a new multi-head attention with grouped fully-connected layers. This yields a compact model with inference speed close to a single network, matching or surpassing Deep Ensembles in UQ performance without retraining from scratch. We also provide an in-depth analysis of pruning, showing that naive approaches can harm calibration, whereas Hydra Ensembles preserves robust uncertainty. Experiments on image and text classification tasks, with various architectures, show consistent gains over Deep Ensembles. Remarkably, in zero-shot classification on ImageNet-1k, our approach surpasses state of the art methods, even without requiring additional training.

cs.LG↗

arXiv2025

Make me an Expert: Distilling from Generalist Black-Box Models into Specialized Models for Semantic Segmentation

The rise of Artificial Intelligence as a Service (AIaaS) democratizes access to pre-trained models via Application Programming Interfaces (APIs), but also raises a fundamental question: how can local models be effectively trained using black-box models that do not expose their weights, training data, or logits, a constraint in which current domain adaptation paradigms are impractical ? To address this challenge, we introduce the Black-Box Distillation (B2D) setting, which enables local model adaptation under realistic constraints: (1) the API model is open-vocabulary and trained on large-scale general-purpose data, and (2) access is limited to one-hot predictions only. We identify that open-vocabulary models exhibit significant sensitivity to input resolution, with different object classes being segmented optimally at different scales, a limitation termed the "curse of resolution". Our method, ATtention-Guided sCaler (ATGC), addresses this challenge by leveraging DINOv2 attention maps to dynamically select optimal scales for black-box model inference. ATGC scores the attention maps with entropy to identify informative scales for pseudo-labelling, enabling effective distillation. Experiments demonstrate substantial improvements under black-box supervision across multiple datasets while requiring only one-hot API predictions. Our code is available at https://github.com/yasserben/ATGC.

cs.CV↗

arXiv2025

Broadband, Flexible, Skin-Compatible Carbon Dots/Graphene Photodetectors for Wearable Applications

The development of wearable photodetectors demands a unique combination of broadband optical sensitivity, mechanical flexibility, and skin-compatibility, with these requirements rarely met simultaneously by existing technologies. Here, we present photodetectors that combine all of these performances. This is achieved by integrating carbon dots, engineered for extended ultraviolet-to-near-infrared absorption, with single-layer graphene transferred onto a plastic substrate. Unlike traditional quantum dot systems, our carbon dots achieve a broad ultraviolet-to-near-infrared response without toxic heavy metals. Graphene provides an efficient channel for charge transport, while a biocompatible chitosan-glycerol electrolyte enables efficient, low-voltage carrier modulation, with peak performance at approximately 0.5 V gate bias. The resulting photodetectors exhibit a broadband photoresponse with responsivities of approximately 0.19 A/W at 406 nm, 0.32 A/W at 642 nm, and 0.18 A/W at 785 nm. They maintain consistent performance at a bending radius of 0.8 cm with negligible degradation after repeated cycles. Furthermore, skin-compatibility assessments using the SkinEthic model confirm the non-toxic nature and suitability of our devices for direct skin contact. The combination of broadband absorption (400-800 nm), flexibility, and skin-compatibility, along with low-voltage operation ($<$ 1.5 V), positions our photodetectors as promising building blocks for next-generation wearable optoelectronics.

cond-mat.other↗

arXiv2025

A Novel Neutrino Mass Matrix

A predictive neutrino mass matrix texture, sheltering unique correlations ($m_{12}=m_{13}\,\,\&\,\,m_{33}=2i\, m_{12}$), is proposed, addressing most of the timely neutrino phenomenology issues. The texture is realized in the framework of type-I + type-II seesaw in the light of the $A_4 \times Z_{10} \times Z_{7} \times Z_{3}$ group. The stability of the proposed texture is studied under renormalization group evolution.

hep-ph↗

arXiv2025

The Echo 12-23 Texture: A Novel Flavour Paradigm for Neutrinos

We construct a flavour guided model that realises the distinctive Echo 12-23 Texture in the neutrino mass matrix through a non-trivial interplay of symmetries. While the model accounts for charged lepton mass hierarchy, the texture offers interesting insights specifically into neutrino mass ordering and flavour dynamics. With clear imprints on low energy observables, the framework provides a minimal yet testable path toward understanding the origin of neutrino properties.

hep-ph↗

arXiv2025

Unification and Texture Universality: The Essence of Hermiticity

A unified framework sheltering the type-I Dirac seesaw, based on the $\Delta(27)$ group supplemented by other cyclic symmetries is proposed, preserving the naturalness of Yukawa couplings. It helps to visualize mass matrices within the lepton and quark sectors in terms of \textit{three universal parameters}: $\Sigma_1$, $\Sigma_2$, and $\Sigma_3$, highlighting that the down-type quark and light neutrino mass matrices exhibit a \textit{Hermitian} texture. Several phenomenological aspects, such as lepton flavour violation and nonunitarity of the lepton mixing matrix are explored. The stability of the proposed texture under renormalization group evolution is also investigated.

hep-ph↗

arXiv2025

FedMVP: Federated Multimodal Visual Prompt Tuning for Vision-Language Models

In federated learning, textual prompt tuning adapts Vision-Language Models (e.g., CLIP) by tuning lightweight input tokens (or prompts) on local client data, while keeping network weights frozen. After training, only the prompts are shared by the clients with the central server for aggregation. However, textual prompt tuning suffers from overfitting to known concepts, limiting its generalizability to unseen concepts. To address this limitation, we propose Multimodal Visual Prompt Tuning (FedMVP) that conditions the prompts on multimodal contextual information - derived from the input image and textual attribute features of a class. At the core of FedMVP is a PromptFormer module that synergistically aligns textual and visual features through a cross-attention mechanism. The dynamically generated multimodal visual prompts are then input to the frozen vision encoder of CLIP, and trained with a combination of CLIP similarity loss and a consistency loss. Extensive evaluation on 20 datasets, spanning three generalization settings, demonstrates that FedMVP not only preserves performance on in-distribution classes and domains, but also displays higher generalizability to unseen classes and domains, surpassing state-of-the-art methods by a notable margin of +1.57% - 2.26%. Code is available at https://github.com/mainaksingha01/FedMVP.

cs.CV↗

arXiv2025

Group-robust Machine Unlearning

Machine unlearning is an emerging paradigm to remove the influence of specific training data (i.e., the forget set) from a model while preserving its knowledge of the rest of the data (i.e., the retain set). Previous approaches assume the forget data to be uniformly distributed from all training datapoints. However, if the data to unlearn is dominant in one group (e.g., ethnicity, gender), we empirically show that performance for this group degrades, leading to fairness issues. To perform unlearning while preserving fairness, this work addresses the overlooked problem of non-uniformly distributed forget sets, which we refer to as group-robust machine unlearning. We formalize the problem and present a simple and effective exact unlearning strategy that mitigates the performance loss in dominant groups via sample distribution reweighting. Moreover, we present MIU (Mutual Information-aware Machine Unlearning), the first approach for group robustness in approximate machine unlearning. MIU minimizes the mutual information between model features and group information, achieving unlearning while reducing performance degradation in the dominant group of the forget set. Additionally, MIU exploits sample distribution reweighting and mutual information calibration with the original model to preserve group robustness. We conduct experiments on three datasets and show that MIU outperforms standard methods, achieving unlearning without compromising model robustness. Source code available at https://github.com/tdemin16/group-robust_machine_unlearning

cs.LG↗

arXiv2025

Permuted Charged Lepton Correction in the Framework of Dirac Seesaw

A Dirac neutrino mass model is proposed, based on an extended group structure of $SU(2)_L \otimes U(1)_Y \otimes A_4 \otimes Z_3 \otimes Z_{10}$ with the Type-I seesaw mechanism. This work explores the impact of parametrization and permutation in the charged lepton diagonalizing matrix, driven by free parameters in the charged lepton sector, on the predictions of observable parameters. Some interesting consequences on the neutrino mass hierarchies, mixing angles and the Dirac CP phase are observed. The framework also finds application in the study of charged lepton flavour violation and dark matter.

hep-ph↗

arXiv2024

Organizing Unstructured Image Collections using Natural Language

In this work, we introduce and study the novel task of Open-ended Semantic Multiple Clustering (OpenSMC). Given a large, unstructured image collection, the goal is to automatically discover several, diverse semantic clustering criteria (e.g., Activity or Location) from the images, and subsequently organize them according to the discovered criteria, without requiring any human input. Our framework, X-Cluster: eXploratory Clustering, treats text as a reasoning proxy: it concurrently scans the entire image collection, proposes candidate criteria in natural language, and groups images into meaningful clusters per criterion. This radically differs from previous works, which either assume predefined clustering criteria or fixed cluster counts. To evaluate X-Cluster, we create two new benchmarks, COCO-4C and Food-4C, each annotated with four distinct grouping criteria and corresponding cluster labels. Experiments show that X-Cluster can effectively reveal meaningful partitions on several datasets. Finally, we use X-Cluster to achieve various real-world applications, including uncovering hidden biases in text-to-image (T2I) generative models and analyzing image virality on social media. Project page: https://oatmealliu.github.io/xcluster.html

cs.CV↗