arXiv subjects

Shreyansh Jain

Publications and source records attributed to Shreyansh Jain.

6 recordsLinked to original sources

arXiv2026

Mathematical Framework for Custom Reward Functions in Job Application Evaluation using Reinforcement Learning

Most of the traditional Applicant Tracking Systems (ATS) depend on strict matching using keywords, where candidates that are highly qualified are many times disqualified because of minor semantic differences. In this article, the two-stage process of developing a more comprehensive resume assessment system based on a small language model that is trained with fewer than 600M parameters is introduced and fine-tuned by using GRPO with a uniquely designed reward function. The initial stage is Supervised Fine-Tuning (SFT), which is used to create a strong base model with the ability to perceive resumes beyond superficial overlap of keywords. This SFT model is further optimized in the second step with Reinforcement Learning (RL) via GRPO with the help of multi-component-based rewarding, which will not be considered as a commission of tokens matching. In the initial RL experiments, we found a severe difficulty in the shape of reward hacking: overly aggressive penalty terms resulted in unstable training dynamics and prohibitively negative model behavior. This was solved by trial-and-error refinement of the reward and careful training hyperparameter tuning, which led to a stable and controlled process of gentle polishing. The GRPO-refined model shows high real-life performance, as it shows an accuracy of 91% on unseen data used for testing. It has a high recall of 0.85 on the SELECTED class with a perfect precision of 1.0, which highlights its high reliability for identifying qualified applicants. These findings demonstrate that an appropriately structured two-step fine-tuning pipeline can effectively be used to transfer a small language model into human-like candidate evaluation, surpassing the shortcomings of both traditional ATS systems and unrefined uses of reinforcement learning.

cs.LG↗

arXiv2025

Noninvasive rheological inference from stable flows in confined tissues

Quantifying the in-plane rheology of epithelial monolayers remains challenging due to the difficulty of imposing controlled shear. We introduce a self-driven, rheometer-like assay in which collective migration generates stationary shear flows, allowing rheological parameters to be inferred directly from image sequences. The assay relies on two sets of ring-shaped fibronectin patches, micropatterned in arrays for high-throughput imaging. Within isolated rings, the epithelial tissue exhibits persistent rotation, from which we infer active migration stresses and substrate friction. Within partially overlapping rings, the tissue exhibits sustained shear, from which we infer the elastic and viscous responses of the cells. The emergence of a Maxwell-like viscoelastic relation -- characterized by a linear relationship between mean cell deformation and neighbor-exchange rate -- is specifically recapitulated within a wet vertex-model framework, which reproduces experimental measurements only when intercellular viscous dissipation is included alongside substrate friction. We apply our method to discriminate the respective roles of two myosin~II isoforms in tissue mechanics. Overall, by harnessing self-generated stresses instead of externally imposed ones, we propose a noninvasive route to rheological inference in migrating epithelial tissues and, more generally, in actively flowing granular materials.

cond-mat.soft↗

arXiv2025

Federated Learning for Commercial Image Sources

Federated Learning is a collaborative machine learning paradigm that enables multiple clients to learn a global model without exposing their data to each other. Consequently, it provides a secure learning platform with privacy-preserving capabilities. This paper introduces a new dataset containing 23,326 images collected from eight different commercial sources and classified into 31 categories, similar to the Office-31 dataset. To the best of our knowledge, this is the first image classification dataset specifically designed for Federated Learning. We also propose two new Federated Learning algorithms, namely Fed-Cyclic and Fed-Star. In Fed-Cyclic, a client receives weights from its previous client, updates them through local training, and passes them to the next client, thus forming a cyclic topology. In Fed-Star, a client receives weights from all other clients, updates its local weights through pre-aggregation (to address statistical heterogeneity) and local training, and sends its updated local weights to all other clients, thus forming a star-like topology. Our experiments reveal that both algorithms perform better than existing baselines on our newly introduced dataset.

cs.CV↗

arXiv2022

Error Correction in ASR using Sequence-to-Sequence Models

Post-editing in Automatic Speech Recognition (ASR) entails automatically correcting common and systematic errors produced by the ASR system. The outputs of an ASR system are largely prone to phonetic and spelling errors. In this paper, we propose to use a powerful pre-trained sequence-to-sequence model, BART, further adaptively trained to serve as a denoising model, to correct errors of such types. The adaptive training is performed on an augmented dataset obtained by synthetically inducing errors as well as by incorporating actual errors from an existing ASR system. We also propose a simple approach to rescore the outputs using word level alignments. Experimental results on accented speech data demonstrate that our strategy effectively rectifies a significant number of ASR errors and produces improved WER results when compared against a competitive baseline. We also highlight a negative result obtained on the related grammatical error correction task in Hindi language showing the limitation in capturing wider context by our proposed model.

cs.CL↗

arXiv2020

Clustering and ordering in cell assemblies with generic asymmetric aligning interactions

Collective cell migration plays an essential role in various biological processes, such as development or cancer proliferation. While cell-cell interactions are clearly key determinants of collective cell migration -- in addition to individual cells self-propulsion -- the physical mechanisms that control the emergence of cell clustering and collective cell migration are still poorly understood. In particular, observations have shown that binary cell-cell collisions generally lead to anti-alignement of cell polarities and separation of pairs -- a process called contact inhibition of locomotion (CIL), which is expected to disfavor the formation of large scale cell clusters with coherent motion. Here, we adopt a joint experimental and theoretical approach to determine the large scale dynamics of cell assemblies from elementary pairwise cell-cell interaction rules. We quantify experimentally binary cell-cell interactions and show that they can be captured by a minimal equilibrium-like pairwise asymmetric aligning interaction potential that reproduces the CIL phenomenology. We identify its symmetry class, build the corresponding active hydrodynamic theory and show on general grounds that such asymmetric aligning interaction destroys large scale clustering and ordering, leading instead to a liquid-like microphase of cell clusters of finite size and short lived polarity, or to a fully dispersed isotropic phase. Finally, this shows that CIL-like asymmetric interactions in cellular systems -- or general active systems -- control cluster sizes and polarity, and can prevent large scale coarsening and long range polarity, except in the singular regime of dense confluent systems.

physics.bio-ph↗

arXiv2019

Python vs. R: A Text Mining Approach for analyzing the Research Trends in Scopus Database

In the contemporary world, with the incubation of advanced technologies and tremendous outbursts of research works, analyzing big data to incorporate research strategies becomes more helpful using the tools and techniques presented in the current research scenario. This paper indeed tries to tackle the most prominent challenges relating to big data analysis by utilizing a text mining approach to analyze research data published in the field of production management as a case to begin with. The study has been conducted by considering research data of International Journal of Production Research (IJPR) indexed in Scopus between 1961-2017 by dividing the analysis incurred into 3 fragments being 1961-1990, 1991-2010 and finally 2011-2017 as a case to highlight the focus of journal. This has indeed provided multi-faceted benefits such as increasing the effectiveness of the procured data with well-established comparisons between R and Python Programming along with providing detailed research trends on the research work incubated. The results of the study highlighted some most prominent topics in the existing IJPR literature such as system's optimization, supplier selection, process design, etc. providing well-established details relating to ongoing research works. The study also compared both languages suiting to a particular field of study for better comprehension and vastness of the research topics. The current research work is one of the part of a copyright work with registration number SW-10310/2018 titled Program for Analyzing Key Trends in Research Data-set. It has been designed in Python for carrying out detailed content analysis based on the available research database in bib format as in the current context it has been applied for IJPR journal and can be replicated on articles of any domain found using keyword search.

cs.CY↗