Extracted assistant, human-compassion and animal-compassion persona vectors from Qwen3-32B served with vLLM on 4xL40S GPUs. Projected activations onto these axes and quantified speciesism between human and animal compassion.
@misc{gupta2026speciesism,title={Speciesism in the Assistant Axis: Probing Compassion Vectors in Post-Trained LLMs},author={Gupta, Shubham and Brazilek, Jasmine},year={2026},howpublished={Poster, MLSys Young Professionals Symposium},}
Inference optimization for model deployment on edge devices. Explores methods such as custom KV caching, speculative decoding, mixed-precision quantization, and quantization-aware training, and how they increase accuracy and token generation throughput for edge LLMs.
@misc{gupta2025edgellms,title={Running Large Models on Small (Edge) Devices},author={Gupta, Shubham},year={2025},howpublished={Poster, Stanford Graph Learning Workshop},}
Machine learning models used to find similarity in large self-collected social media username datasets.
@inproceedings{gupta2020identitylinkage,title={Investigation of Biases in Identity Linkage Datasets},author={Kaushal, Rishabh and Gupta, Shubham and Kumaraguru, Ponnurangam},booktitle={Proceedings of the 35th Annual ACM Symposium on Applied Computing (SAC '20)},pages={1861--1868},year={2020},publisher={ACM},doi={10.1145/3341105.3374015},}
A decentralized reputation system for crowdsourcing platforms built on the Ethereum blockchain.
@misc{gupta2020workerrep,title={WorkerRep: Immutable Reputation System for Crowdsourcing Platform Based on Blockchain},author={Bhatia, Gurpriya Kaur and Gupta, Shubham and Dubey, Alpana and Kumaraguru, Ponnurangam},year={2020},archiveprefix={arXiv},doi={10.48550/arXiv.2006.14782},}
Awarded best paper in the CMU Network Security class.
A survey of typo-squatting attacks across software development ecosystems, done as a course project.
@misc{gupta2020typosquatting,title={A Survey of Typo-Squatting Attacks in Development Ecosystems},author={Gupta, Shubham},year={2020},howpublished={Course project, Network Security, Carnegie Mellon University},}
A cross-consistency tool providing periodic updates about discrepancies between two CMS metadata databases, DBS and PhEDEx.
@misc{gupta2016cmsmetadata,title={Database Consistency Verification for CMS Metadata},author={Gupta, Shubham and Kuznetsov, Valentin and Dziedziniewicz-Wojcik, Katarzyna Maria and Magini, Nicolò},year={2016},howpublished={CERN Openlab report, Zenodo},doi={10.5281/zenodo.61230},}