BZPEERPapers, people, evidence and research paths
Menu

Discover

SearchFind papers, people, topics, institutions and opportunities from one place.GraphSee how papers, authors, topics and evidence connect.

Research tools

HomeContinue from the papers, people and topics that matter now.CollectionsTurn saved papers into reading lists, evidence maps and shareable context.ResearchReconstruct the path that led you from search to paper to question.CollaborationMove from a paper, topic or expert request into a focused thread.PublishCreate full-text research articles with metadata, formulas and versions.

Network

ExpertsFind specialists whose work and reviews explain why they can help.CommunitiesJoin research groups organized around papers, topics and shared evidence.

Opportunities

ListingsBrowse roles, calls and collaborations connected to real research context.

Account

Log inReturn to your saved papers, graph views and active threads.Create accountStart saving papers, building a research trail and joining teams.
Log inCreate account

Source author record

Subhankar Roy

Subhankar Roy appears in the imported research catalog. Authorship, coauthor and topic links are available while profile ownership is still unclaimed.

ResearcherUnclaimed source record
hep-phComputer VisionMachine LearningArtificial IntelligenceComputational Engineering, Finance, and ScienceQuantitative Methods

Catalog footprint

What is connected

14works
6topics
4close collaborators

Actions

Connect this record

Log in to claim
Open graphBrowse works

Research graph

See the researcher in context

Open full explorer

Inspect adjacent papers, topics, institutions and collaborators without losing the researcher page.

Building this map preview

BZPEER is loading the nearby papers, people, topics and institutions for this page.

Published work

14 published item(s)

preprint2026arXiv

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.

0 citations0 reviewsNo code linkedOpen access
preprint2023arXiv

Simplifying Open-Set Video Domain Adaptation with Contrastive Learning

In an effort to reduce annotation costs in action recognition, unsupervised video domain adaptation methods have been proposed that aim to adapt a predictive model from a labelled dataset (i.e., source domain) to an unlabelled dataset (i.e., target domain). In this work we address a more realistic scenario, called open-set video domain adaptation (OUVDA), where the target dataset contains "unknown" semantic categories that are not shared with the source. The challenge lies in aligning the shared classes of the two domains while separating the shared classes from the unknown ones. In this work we propose to address OUVDA with an unified contrastive learning framework that learns discriminative and well-clustered features. We also propose a video-oriented temporal contrastive loss that enables our method to better cluster the feature space by exploiting the freely available temporal information in video data. We show that discriminative feature space facilitates better separation of the unknown classes, and thereby allows us to use a simple similarity based score to identify them. We conduct thorough experimental evaluation on multiple OUVDA benchmarks and show the effectiveness of our proposed method against the prior art.

0 citations0 reviewsNo code linkedOpen access
preprint2022arXiv

Class-incremental Novel Class Discovery

We study the new task of class-incremental Novel Class Discovery (class-iNCD), which refers to the problem of discovering novel categories in an unlabelled data set by leveraging a pre-trained model that has been trained on a labelled data set containing disjoint yet related categories. Apart from discovering novel classes, we also aim at preserving the ability of the model to recognize previously seen base categories. Inspired by rehearsal-based incremental learning methods, in this paper we propose a novel approach for class-iNCD which prevents forgetting of past information about the base classes by jointly exploiting base class feature prototypes and feature-level knowledge distillation. We also propose a self-training clustering strategy that simultaneously clusters novel categories and trains a joint classifier for both the base and novel classes. This makes our method able to operate in a class-incremental setting. Our experiments, conducted on three common benchmarks, demonstrate that our method significantly outperforms state-of-the-art approaches. Code is available at https://github.com/OatmealLiu/class-iNCD

0 citations0 reviewsNo code linkedOpen access
preprint2022arXiv

Uncertainty-guided Source-free Domain Adaptation

Source-free domain adaptation (SFDA) aims to adapt a classifier to an unlabelled target data set by only using a pre-trained source model. However, the absence of the source data and the domain shift makes the predictions on the target data unreliable. We propose quantifying the uncertainty in the source model predictions and utilizing it to guide the target adaptation. For this, we construct a probabilistic source model by incorporating priors on the network parameters inducing a distribution over the model predictions. Uncertainties are estimated by employing a Laplace approximation and incorporated to identify target data points that do not lie in the source manifold and to down-weight them when maximizing the mutual information on the target data. Unlike recent works, our probabilistic treatment is computationally lightweight, decouples source training and target adaptation, and requires no specialized source training or changes of the model architecture. We show the advantages of uncertainty-guided SFDA over traditional SFDA in the closed-set and open-set settings and provide empirical evidence that our approach is more robust to strong domain shifts even without tuning.

0 citations0 reviewsNo code linkedOpen access
preprint2021arXiv

Metric-Learning based Deep Hashing Network for Content Based Retrieval of Remote Sensing Images

Hashing methods have been recently found very effective in retrieval of remote sensing (RS) images due to their computational efficiency and fast search speed. The traditional hashing methods in RS usually exploit hand-crafted features to learn hash functions to obtain binary codes, which can be insufficient to optimally represent the information content of RS images. To overcome this problem, in this paper we introduce a metric-learning based hashing network, which learns: 1) a semantic-based metric space for effective feature representation; and 2) compact binary hash codes for fast archive search. Our network considers an interplay of multiple loss functions that allows to jointly learn a metric based semantic space facilitating similar images to be clustered together in that target space and at the same time producing compact final activations that lose negligible information when binarized. Experiments carried out on two benchmark RS archives point out that the proposed network significantly improves the retrieval performance under the same retrieval time when compared to the state-of-the-art hashing methods in RS.

0 citations0 reviewsNo code linkedOpen access
preprint2020arXiv

Unsupervised Domain Adaptation using Feature-Whitening and Consensus Loss

A classifier trained on a dataset seldom works on other datasets obtained under different conditions due to domain shift. This problem is commonly addressed by domain adaptation methods. In this work we introduce a novel deep learning framework which unifies different paradigms in unsupervised domain adaptation. Specifically, we propose domain alignment layers which implement feature whitening for the purpose of matching source and target feature distributions. Additionally, we leverage the unlabeled target data by proposing the Min-Entropy Consensus loss, which regularizes training while avoiding the adoption of many user-defined hyper-parameters. We report results on publicly available datasets, considering both digit classification and object recognition tasks. We show that, in most of our experiments, our approach improves upon previous methods, setting new state-of-the-art performances.

0 citations0 reviewsNo code linkedOpen access
preprint2016arXiv

Generating nonzero $θ_{13}$ without breaking the $2$-$3$ symmetry of neutrino mass matrix

The prediction of vanishing reactor angle was thought to be a signature of $2$-$3$ symmetry of neutrino mass matrix. But the present study addresses certain interesting facts related with $2$-$3$ symmetry which are not addressed so far. The investigation highlights that $θ_{13}=0$, corresponds to a very special case in association with $2$-$3$ symmetry and to engender a non-zero $θ_{13}$, the breakdown of $2$-$3$ symmetry is not essential.

0 citations0 reviewsNo code linkedOpen access
preprint2016arXiv

Modulated bimaximal neutrino mixing

The present article is an endeavor to look into some fruitful frameworks based on "Bi-maximal" neutrino mixing, from a model independent stand. The possibilities involving the correction or attenuation of the original BM mixing matrix, followed by GUT-inspired charged lepton correction are invoked. The "symmetry-basis" thus constructed, accentuates some interesting facets such as: a modified QLC relation, $θ_{12}+θ_{c}\approx\fracπ{4}-θ_{13}\cos(nπ-δ_{CP})$, a possible link up between neutrino and charged lepton sectors, $θ_{13}^ν=θ_{12}^{l}\sim\mathcal{O}(θ_{C})$ or that between neutrinos and quarks, $θ_{13}^ν=θ_{C}$. The study vindicates the relevance of the Bi-maximal mixing as a first approximation.

0 citations0 reviewsNo code linkedOpen access
preprint2016arXiv

The mixing angle as a function of neutrino mass ratio

In the quark sector, we experience a correlation between the mixing angles and the mass ratios. A partial realization of the similar tie-up in the neutrino sector helps to constrain the parametrization of masses and mixing, and hints for a predictive framework. We derive five hierarchy dependent textures of neutrino mass matrix with minimum number of parameters ($\leq\,4$), following a model-independent strategy.

0 citations0 reviewsNo code linkedOpen access
preprint2013arXiv

Leptonic mixing matrix in terms of Cabibbo angle

We phenomenologically build a neutrino mass matrix obeying $μ-τ$ symmetry with only two parameters, Cabibbo angle $(λ)$ and a flavour twister parameter $(η)$. For vanishing $η$, the model assumes TBM mixing. When $η=λ$, the model generates a solar mixing angle and solar mass squared difference that coincide with the experimental best-fits. It motivates us to propose a mixing matrix in the neutrino sector with $ sinθ_{12}<\frac{1}{\sqrt{3}}$, $θ_{13}=0$ and $θ_{23}=π/4$. The corrections in $θ_{13}$ and $θ_{23}$ are conducted following the breaking of $μ-τ$ symmetry, by choosing a proper unitary diagonalizing charged lepton matrix, and this ensures that $θ_{23}$ lies within the first octant. The Cabibbo angle $(λ=\sinθ_{c})$ plays the role of a guiding parameter in both neutrino as well as leptonic sector. The effect of the Dirac CP violating phase $δ_{cp}$ is also studied when it enters either through the charged lepton diagonalizing matrix or through neutrino mixing matrix.

0 citations0 reviewsNo code linkedOpen access
preprint2012arXiv

A new method of parametrisation of neutrino mass matrix through breaking of $μ-τ$ symmetry: Normal hierarchy

In the first part of the present work the $μ-τ$ symmetry of the neutrino mass matrix is perturbed at its minimal level in order to produce deviation from Tri-bimaximal mixing (TBM), which includes nonzero value of reactor angle $θ_{13}$ and maximal condition of $\tan^{2}θ_{23}=1$. The parametrisation of neutrino mass matrix which describes Normal hierarchy (NH), has been addressed with minimum number of independent parameters, out of which two parameters $η$ and $α$ take care of $θ_{12}$ and $θ_{13}$ respectively without any interference with mass eigenvalues. In the second part the deviation from maximal condition $\tan^{2}θ_{23}=1$, along with the nonzero value of $θ_{13}$, has been implemented with the introduction of a perturbing matrix which breaks the $μ-τ$ symmetric mass matrix. The model is found to be flexible enough to adjust itself with the changing precise experimental results. The method is also applicable for inverted hierarchy and quasi-degenerate cases.

0 citations0 reviewsNo code linkedOpen access
preprint2012arXiv

An Efficient Biological Sequence Compression Technique Using LUT And Repeat In The Sequence

Data compression plays an important role to deal with high volumes of DNA sequences in the field of Bioinformatics. Again data compression techniques directly affect the alignment of DNA sequences. So the time needed to decompress a compressed sequence has to be given equal priorities as with compression ratio. This article contains first introduction then a brief review of different biological sequence compression after that my proposed work then our two improved Biological sequence compression algorithms after that result followed by conclusion and discussion, future scope and finally references. These algorithms gain a very good compression factor with higher saving percentage and less time for compression and decompression than the previous Biological Sequence compression algorithms. Keywords: Hash map table, Tandem repeats, compression factor, compression time, saving percentage, compression, decompression process.

0 citations0 reviewsNo code linkedOpen access
preprint2012arXiv

Bi-Large neutrino mixing with charged lepton correction

The usual Bi-Maximal (BM) neutrino mixing faces an inherent problem in lowering the solar mixing angle below $\tan^{2}θ_{12}=0.50$ when charged lepton correction is taken. This minimum $θ_{12}$ is achievable only if CP violation is absent. We start with a new model which incorporates a new idea of mixing developed recently, called Bi-Large (BL) mixing, similar to BM mixing execpt that the former chooses rather $θ_{13}$ as Cabibbo angle $(θ_c)$ than zero. We apply this mixing in the neutrino sector, followed by a charged lepton correction with the CKM type matrix $U_{l}$. This model marks a prediction on $θ_{23}$ to lie within the first octant. The CP violating phase $δ_{CP}$ dictates the prediction of all the three mixing angles. A proper choice of $δ_{CP}$ leads to the predictions of all the three mixing angles including $θ_{12}$, to align very precisely with the experimental bestfits. This close agreement thus hoists Bi-Large mixing as an important and promising mixing scheme, in contrast to BM or TBM mixing as a first approximation. A formal derivation of BL mixing from discrete symmetry will be an important investigation in neutrino physics.

0 citations0 reviewsNo code linkedOpen access
preprint2012arXiv

Expansion of $U_{PMNS}$ and Neutrino mass matrix $M_ν$ in terms of $sinθ_{13}$ for Inverted Hierarchical case

The recent observational data supports the deviation from Tri-bimaximal (TBM) mixings. Different neutrino mass models suggest the interdependency among the observational parameters involving the mixing angles. On phenomenological ground we try to construct the PMNS matrix $U_{PMNS}$ with certain analytic structure satisfying the unitary condition, in terms of a single observational parameter $sinθ_{13}$. We hypothesise the three neutrino masses $m_{i}$ as functions of $sinθ_{13}$ and then construct the neutrino mass matrix $M_ν$. We assume the convergence of the model to TBM mixing when $θ_{13}$ is taken zero. This mass matrix so far obtained can be employed for various applications including the estimation of matter-antimatter asymmetry of the Universe.

0 citations0 reviewsNo code linkedOpen access