Nitish Shukla

2335, Engineering Building

Michigan State University

East Lansing, MI, 48864

I am a Ph.D. student in Computer Science at Michigan State University, advised by Prof. Arun Ross in the iPRoBe Lab. My research focuses on efficient, interpretable, and reliable visual AI, spanning multimodal learning, generative modeling, and biometric security.

A central question connects my work: How can we use the information within pretrained models to make visual inference more efficient and dependable? I investigate how models represent visual evidence, how that evidence can guide decisions under computational constraints, and how generative models can recover identity information from ambiguous images.

My work on facial demorphing provides a concrete foundation for this research. A morphed face combines two identities into a single image; recovering the contributing identities requires more than producing visually plausible reconstructions. It requires models that preserve identity, generalize to unfamiliar faces and morphing methods, and are evaluated against meaningful biometric criteria. These challenges also motivate my broader work on multimodal representations and reliable inference.

Research directions

Efficient multimodal inference: selecting visual evidence across images and model layers to support reasoning within an explicit computational budget.
Multimodal representation analysis: studying intermediate model representations for interpretable detection and attribute-specific image retrieval.
Generative inverse problems: developing diffusion models, GANs, and latent optimization methods to recover constituent identities from a single morphed face.
Trustworthy visual AI: investigating adversarial robustness, morphing attack detection, and evaluation protocols that distinguish identity recovery from reconstruction shortcuts.

Read more about the questions and methods on my research page, explore individual projects, or see my publications.

Research highlights

Findings of ACL 2026: S2H-DPO, a hardness-aware preference optimization framework for vision–language models.
IAPR Best Biometrics Student Paper Award, IJCB 2025: facial demorphing from a single morph using a latent conditional GAN.
Adobe Research internships, 2025 and 2026: research on attribute-specific image retrieval and robust enterprise Text-to-SQL.

Background

My training in mathematics and computer science informs my approach to representation learning and model evaluation. Before joining MSU, I earned an M.Sc. in Computer Science from Chennai Mathematical Institute and an M.Sc. in Mathematics from IIT Guwahati. At CMI, I worked with Prof. K. V. Subramanyam and Prof. Madhavan Mukund on robustness properties induced by prototypes in deep neural networks.

Outside research, I enjoy cooking, films, video games, and hiking, especially in snowy mountain landscapes. I share some of my cooking on Instagram. For research discussions and collaboration, please contact me.

news

Apr 07, 2026	Paper titled “S2H-DPO: Hardness-Aware Preference Optimization for Vision–Language Models” has been accepted to ACL 2026!. 🎊
Sep 11, 2025	Awarded IAPR Best Biometrics Student Paper Award (BBSPA) at IEEE International Joint Conference on Biometrics (IJCB), 2025. 🎊
Jul 12, 2025	Paper titled “Facial Demorphing from a Single Morph Using a Latent Conditional GAN” accepted at IEEE International Joint Conference on Biometrics (IJCB), 2025.
May 20, 2025	Paper titled “diffDemorph: Extending Reference-Free Demorphing to Unseen Faces” accepted at IEEE International Conference on Image Processing (ICIP), 2025.
May 12, 2025	Started summer internship at Adobe Research. 🎊

selected publications

arXiv
Enhancing Single-Image Facial Demorphing using Multimodal Large Language Models

Nitish Shukla, and Arun Ross

arXiv preprint arXiv:2605.25442, 2026

Abs arXiv Bib

A reference-free facial demorphing framework that leverages Multimodal Large Language Models (MLLMs) to guide a coupled diffusion-based reconstruction. Semantic embeddings extracted from intermediate MLLM layers condition the demorphing, providing high-level reasoning about facial attributes and identity cues that complement low-level pixel information. Both constituent faces are synthesized jointly through a denoising diffusion model operating directly in the RGB domain. The method achieves restoration accuracies exceeding 96% at 0.1% FMR on landmark-based morphs, with PSNR gains of 6–9 dB over existing methods.
@article{shukla2026llmdemorph, title = {Enhancing Single-Image Facial Demorphing using Multimodal Large Language Models}, author = {Shukla, Nitish and Ross, Arun}, journal = {arXiv preprint arXiv:2605.25442}, year = {2026}, }
ACL
S2H-DPO: Hardness-Aware Preference Optimization for Vision–Language Models

Nitish Shukla, Surgan Jandial, and Arun Ross

In Findings of the Association for Computational Linguistics (ACL), 2026

Abs Bib HTML

A Simple-to-Hard (S2H) learning framework that systematically constructs multi-image preference data across three hierarchical reasoning levels: single-image localized reasoning, multi-image localized comparison, and global visual search. Unlike prior work relying on model-specific attributes such as hallucinations or attention heuristics, the approach leverages prompt-driven complexity to create chosen/rejected pairs applicable across different models. Evaluations on LLaVA and Qwen-VL show significant gains in multi-image reasoning while preserving single-image performance.
@inproceedings{shukla2026s2hdpo, title = {S2H-DPO: Hardness-Aware Preference Optimization for Vision--Language Models}, author = {Shukla, Nitish and Jandial, Surgan and Ross, Arun}, booktitle = {Findings of the Association for Computational Linguistics (ACL)}, year = {2026}, }
IJCB
Facial Demorphing from a Single Morph Using a Latent Conditional GAN

Nitish Shukla, and Arun Ross

In IEEE International Joint Conference on Biometrics (IJCB), 2025

Awarded Abs arXiv Bib Code

IAPR Best Biometrics Student Paper Award (BBSPA)

The method decomposes a morph in latent space, allowing it to demorph images created from unseen morph techniques and face styles. Trained on morphs created from synthetic faces and tested on morphs created from real faces using different morph techniques, the method outperforms existing methods by a considerable margin and produces high-fidelity demorphed face images.
@inproceedings{shukla2025lcgan, title = {Facial Demorphing from a Single Morph Using a Latent Conditional GAN}, author = {Shukla, Nitish and Ross, Arun}, booktitle = {IEEE International Joint Conference on Biometrics (IJCB)}, year = {2025}, }
ICIP
diffDeMorph: Extending Reference-Free Demorphing to Unseen Faces

Nitish Shukla, and Arun Ross

In IEEE International Conference on Image Processing (ICIP), 2025

Abs arXiv Bib

A diffusion-based approach, diffDeMorph, that disentangles component images from a composite morph with high visual fidelity. It is the first to generalize across morph techniques and face styles, beating the prior state of the art by at least 59.46% under a common training protocol across all datasets tested. The model is trained on morphs from synthetically generated faces and tested on real morphs.
@inproceedings{shukla2025diffdemorph, title = {diffDeMorph: Extending Reference-Free Demorphing to Unseen Faces}, author = {Shukla, Nitish and Ross, Arun}, booktitle = {IEEE International Conference on Image Processing (ICIP)}, year = {2025}, }