Dr. Siladittya Manna
I investigate vulnerabilities in PEFT-based federated learning frameworks and develop trustworthy distributed AI systems.
About Me
“Building trustworthy machine learning systems that learn collaboratively while preserving privacy, security, and sample efficiency.”
- Federated Learning
- PEFT Security
- Gradient Inversion
- Self-Supervised Learning
- Representation Learning
- Medical AI
- Privacy-Preserving AI
Who I Am
I am an ANRF Postdoctoral Fellow at the Indian Institute of Science (IISc), Bangalore, working in the Department of Computational and Data Sciences with Prof. Anirban Chakraborty. Prior to IISc, I was a Senior Research Assistant at Hong Kong Baptist University with Prof. Yiu-Ming Cheung, and completed my Ph.D. in Computer Science at the Indian Statistical Institute (ISI), Kolkata, under the supervision of Prof. Umapada Pal.Research Vision
My overarching goal is to build privacy-preserving, adversarial-resilient distributed AI systems. When healthcare providers or autonomous agents collaborate across institutions, sharing parameters must not compromise proprietary data. I investigate how parameter-efficient adapters (PEFT) and foundation models can be secured against inversion attacks, poisoning, and information leakage without sacrificing convergence speed.Research Trajectory (How the pieces connect)
- Foundation (Ph.D. at ISI Kolkata)Self-supervised representation learning and label-efficient medical image analysis.
- Distributed Expansion (HKBU & IISc)Federated learning across decentralized medical data with non-IID heterogeneity.
- Current Frontier (Postdoc at IISc)AI security, gradient inversion attacks, and privacy vulnerabilities in parameter-efficient fine-tuning (PEFT) adapters.
Latest News
Breaking the Privacy Promise: Gradient Inversion Attacks on Federated Low-Rank Adapted Vision Transformers
Congratulations to Ritik, Ashmit, Yashwanth and Prof. Anirban Chakraborty, for their paper being accepted at the Winter Conference on Applications of Computer Vision 2027 (WACV 2027).
ANRF National Post-Doctoral Fellowship (2026 to 2028)
Awarded the prestigious ANRF National Post-Doctoral Fellowship by the Anusandhan National Research Foundation (DST, Govt. of India). Hosted at the Department of CDS, IISc Bangalore, under the mentorship of Prof. Anirban Chakraborty.
Reliability-Aware Spatio-Temporal Maps for Heart Rate Monitoring (ICIP 2026)
Co-authored with Arpan Bairagi, Rakesh Dey, and Prof. Umapada Pal (ISI Kolkata). Accepted and to be presented at IEEE International Conference on Image Processing (ICIP 2026).
Semi-Supervised Sperm Motility Classification on VISEM-Tracking (AI4Fertility 2026)
Co-authored with Suyash Kumar, Ankur Singh (IIT BHU), and Dr. Rajkumar Saini (LUT University, Sweden). To be presented at The First Workshop on AI in Fertility Science 2026.
Residual Dense Blocks for Foreground Imbalance in Brachytherapy Applicator Segmentation
Published in Springer Nature Computer Science. Co-authored with Suresh Das, Prasun Sanki, Subhayan Mondal, Dr. Saumik Bhattacharya, and Dr. Sayantari Ghosh.
Selected Publications
Peer-reviewed journal articles, conference proceedings, survey papers, and doctoral research in Self-Supervised Learning and Medical AI.
Showing 20 publications
- 2027
Breaking the Privacy Promise: Gradient Inversion Attacks on Federated Low-Rank Adapted Vision Transformers
We investigate the sparsely explored domain of the risk of private data leakage in LoRA-adapted Vision Transformers (ViTs), where low-rank updates weaken the gradient signal available to an honest-but-curious server.
- Medical AI
- Federated Learning
- 2026
Bridging the Applicator Gap with Data-Doping: Dual-Domain Learning for Precise Bladder Segmentation in CT-Guided Brachytherapy
This study highlights the value of integrating anatomically similar but distribution shifted datasets to overcome data scarcity and enhance deep learning based segmentation for brachytherapy treatment planning.
- Medical AI
- Deep Learning
Preprint
- 2026
On the Role of the Projector in Contrastive Self-Supervised Learning: Last-Layer Rank Dynamics Drive Representation Quality
Our method improves Top-1 accuracy by more than 1% on SimCLR on the ImageNet100 dataset and consistently outperforms baseline SimCLR variants on CIFAR datasets, supporting our interpretation of the projector's role.
- Self-Supervised Learning
Preprint
- 2026
Multiorgan Segmentation in Magnetic Resonance Images for Gynecological Brachytherapy Using Residuals in Residual Dense Block
RU-Net offers a robust solution for multiorgan segmentation in gynecological BT.
- Medical AI
- Deep Learning
- 2026
Reliability-Aware Weighted Multi-Scale Spatio-Temporal Maps for Heart Rate Monitoring
A novel reliability-aware multi-scale spatio-temporal map representation for contact-free heart rate monitoring, improving rPPG robustness against motion artifacts.
- Medical AI
- Representation Learning
- 2026
Semi-Supervised Sperm Motility Classification Using WHO Kinematic Features and Domain-Adapted Detection on VISEM-Tracking
A semi-supervised framework leveraging WHO kinematic criteria and domain adaptation for accurate sperm motility tracking and clinical fertility classification.
- Medical AI
- Self-Supervised Learning
- 2025
Dynamically Scaled Temperature in Self-Supervised Contrastive Learning
We introduce a dynamic temperature scaling mechanism for self-supervised contrastive learning to improve representation quality across varied data distributions.
- Self-Supervised Learning
- Representation Learning
- 2025
Decorrelation-Based Self-Supervised Visual Representation Learning for Writer Identification
We propose a modified formulation of the decorrelation-based framework named SWIS for writer identification by standardizing cross-dimensional features.
- Self-Supervised Learning
- Representation Learning
- 2025
Residual Dense Blocks for Extreme Foreground Imbalance in Brachytherapy Applicator Segmentation
Detecting radioactive probes in gynecological brachytherapy imaging under extreme foreground imbalance using residual dense block architectures.
- Medical AI
- 2025
Self-Supervised Learning and Its Applications in Medical Image Analysis
Ph.D. Doctoral Dissertation exploring foundational self-supervised representation learning and domain-adapted pretext objectives for clinical medical imaging.
- Self-Supervised Learning
- Medical AI
- Representation Learning
- 2024
Self-Supervised Visual Representation Learning for Medical Image Analysis: A Comprehensive Survey
A comprehensive review of self-supervised learning principles and their applications to medical image analysis, compiling performance metrics across major benchmarks.
- Self-Supervised Learning
- Medical AI
- 2024
Correlation Weighted Prototype-based Self-Supervised One-Shot Segmentation of Medical Images
A correlation weighted prototype-based approach for self-supervised one-shot segmentation of medical images with limited expert annotations.
- Self-Supervised Learning
- Medical AI
- Representation Learning
27th International Conference on Pattern Recognition (ICPR 2024)
Conference, Featured
- 2023
Self-supervised representation learning for knee injury diagnosis from magnetic resonance data
A self-supervised learning framework for knee injury diagnosis using magnetic resonance imaging data, alleviating heavy reliance on manual labels.
- Self-Supervised Learning
- Medical AI
- Representation Learning
- 2023
Selfdocseg: A self-supervised vision-based approach towards document segmentation
Selfdocseg presents a self-supervised vision-based framework for document layout segmentation without manual ground-truth annotations.
- Self-Supervised Learning
- Representation Learning
- 2023
MIO: Mutual Information Optimization using Self-Supervised Binary Contrastive Learning
Mutual information optimization using self-supervised binary contrastive learning for sample-efficient representation learning.
- Self-Supervised Learning
- Representation Learning
- 2022
Surds: Self-supervised attention-guided reconstruction and dual triplet loss for writer independent offline signature verification
Attention-guided self-supervised reconstruction combined with dual triplet loss for offline signature verification.
- Self-Supervised Learning
- Representation Learning
26th International Conference on Pattern Recognition (ICPR 2022)
Conference, Featured
- 2022
SWIS: Self-Supervised Representation Learning for Writer Independent Offline Signature Verification
A self-supervised representation learning framework SWIS for writer-independent offline signature verification.
- Self-Supervised Learning
- Representation Learning
- 2022
Self-supervised representation learning for detection of ACL tear injury in knee MR videos
Self-supervised representation learning approach for detecting ACL tear injuries in volumetric knee MR videos.
- Self-Supervised Learning
- Medical AI
- Representation Learning
- 2022
PLSM: A Parallelized Liquid State Machine for Unintentional Action Detection
A parallelized liquid state machine architecture for unintentional action detection in video streams.
- Representation Learning
- 2021
Interpretive self-supervised pre-training: boosting performance on visual medical data
Interpretive self-supervised pre-training techniques to enhance downstream transfer on visual medical datasets.
- Self-Supervised Learning
- Medical AI
- Representation Learning
Indian Conference on Computer Vision, Graphics and Image Processing (ICVGIP)
Conference
Research Areas
Integrated research domains bridging distributed optimization, threat analysis, representation learning, and medical AI.
Federated Learning
Developing resilient distributed learning algorithms that train collaboratively across decentralized, non-IID client partitions with parameter-efficient fine-tuning (PEFT) and adaptive aggregation.
- Distributed Optimization
- PEFT Adapters
- Personalization
- Non-IID Heterogeneity
AI Security & Privacy
Auditing and fortifying learning systems against adversarial threats, including gradient inversion attacks, prompt poisoning in foundation models, and verifiable machine unlearning.
- Gradient Inversion Attacks
- Prompt Poisoning
- Machine Unlearning
- Privacy Guarantees
Self-Supervised Representation Learning
Advancing contrastive and generative pretext objectives that acquire transferable, sample-efficient visual representations from unlabeled and multimodal data.
- Contrastive Learning
- Mutual Information
- Vision Transformers
- Foundation Models
Medical AI & Clinical Diagnostics
Bridging machine learning theory with clinical healthcare, focusing on label-efficient organ segmentation, rare foreground imbalance, and safe diagnostic assistance.
- Medical Image Analysis
- Few-Shot Segmentation
- Brachytherapy Tracking
- Clinical AI
Current Research Questions
- How do parameter-efficient fine-tuning (PEFT) adapters interact with gradient inversion vulnerabilities during federated model synchronization?
- Can decentralized vision-language models maintain robustness against prompt poisoning and adversarial manipulation across heterogeneous clients?
- How can self-supervised representation learning guarantee label-efficiency in clinical imaging without succumbing to severe domain shift?
- How can we reconcile client personalization with formal differential privacy and Byzantine fault tolerance in distributed optimization?
Long-Term Research Goals
- Trustworthy Distributed AI
- Establish distributed learning frameworks whose empirical performance matches centralized baselines while offering verifiable security against gradient leakage.
- Sample-Efficient Representation Learning
- Devise self-supervised pretext tasks that transfer seamlessly across multimodal domains with minimal manual curation.
- Secure Cross-Institutional Collaboration
- Enable hospitals and research institutions to train state-of-the-art shared models without transmitting raw patient data.
- High-Stakes Clinical Translation
- Deploy robust, interpretable computer vision algorithms for real-world clinical interventions in oncology and reproductive health.
Ongoing Projects
Active research investigations, manuscripts under review, and ongoing collaborative studies across Federated Learning security and medical AI.
Gradient Inversion Attacks on Parameter-Efficient Federated Learning
Ongoing, Indian Institute of Science, Bangalore
Evaluating attack surfaces in federated learning.
Collaborators: Ritik Badiya, Ashmit Sinha, Rushikesh Gaikar, Aditi Prabakaran, Yashwanth Mandula, Prof. Anirban Chakraborty
- Federated Learning
- AI Security
- Gradient Inversion
- PEFT Adapters
On the Role of Projector in Self-Supervised Contrastive Learning
Under Review, IIT Kharagpur & ISI Kolkata
Investigating the role of projector in self-supervised contrastive learning and finding the critical reasons behind dimensional collapse.
Collaborators: Priyangshu Mandal, Dr. Saumik Bhattacharya, Prof. Umapada Pal
- Self-Supervised Learning
- Contrastive Learning
- Representation Learning
- Dimensional Collapse
Domain-Adaptive Object Detection
Ongoing, University of Lille, France; Universitat Autònoma de Barcelona, Spain
Developing domain-adaptive object detection methods for visually similar objects by exploiting domain-invariant features and cross-domain alignment strategies.
Collaborators: Dr. Tanmoy Mondal, Tristan Groussard, Dr. Sanket Biswas
- Domain Adaptation
- Object Detection
- Cross-Domain Generalization
- Visually Similar Objects
LLM Training on Low-Resource Language
Ongoing, University of Hildesheim, Germany; Bennett University, India
Adapting Large Language Models for low-resource Indian languages through data-centric approaches and parameter-efficient fine-tuning strategies.
Collaborators: Dr. Koyel Ghosh, Dr. Supriya Chandra
- LLM
- Low-Resource Language
- Parameter-Efficient Fine-Tuning
Team
Colleagues, researchers, and students I am currently collaborating with at the Indian Institute of Science (IISc), Bangalore.
| Name | Role | Affiliation | Focus | Profile |
|---|---|---|---|---|
| Ritik Kumar Badiya | Graduate Researcher (M.Tech) | Visual Computing Lab, CDS, IISc | Gradient Inversion Attacks on PEFT | Profile Link |
| Ashmit Sinha | Research Intern | Visual Computing Lab, CDS, IISc | Gradient Inversion in Federated Learning | Profile Link |
| Aditi Prabakaran | Research Intern | Visual Computing Lab, CDS, IISc | Gradient Inversion Attacks | Profile Link |
| Rushikesh Gaikar | Research Intern | Department of CDS, IISc Bangalore | Gradient Inversion Attacks on PEFT | Profile Link |
| Yashwanth Mandula | Ph.D. Scholar | Visual Computing Lab, CDS, IISc | Federated Learning | Profile Link |
Awards & Funding
ANRF National Post-Doctoral Fellowship (N-PDF)
Anusandhan National Research Foundation (ANRF), Department of Science & Technology, Government of India
- Premier National Fellowship
- 2026 to 2028, Active Award
- Host Institute
- Indian Institute of Science (IISc), Bangalore
- Faculty Mentor
- Prof. Anirban Chakraborty (Dept. of CDS)
- Research Focus
- Security & Privacy in PEFT-based Federated Learning
The ANRF National Post-Doctoral Fellowship is one of the highest individual postdoctoral honors in India, awarded through rigorous national peer-review to support independent, early-career researchers conducting high-impact scientific inquiry.
Journey & Experience
Post-Doctoral Fellow (ANRF)
Indian Institute of Science, Bangalore, India
Advisor: Prof. Anirban Chakraborty
Investigating privacy vulnerabilities in parameter-efficient fine-tuning (PEFT) adapters and distributed multimodal architectures.
Post-Doctoral Research Associate
Indian Institute of Science, Bangalore, India
Advisor: Prof. Anirban Chakraborty
Research on vulnerabilities in Transformer-based federated learning and distributed optimization frameworks.
Senior Research Assistant
Hong Kong Baptist University, Kowloon, Hong Kong
Advisor: Prof. Yiu-Ming Cheung
Conducted research on federated learning algorithms applied to medical image segmentation, augmented by self-supervised learning principles.
Ph.D. in Computer Science
Indian Statistical Institute, Kolkata, India
Advisor: Prof. Umapada Pal
Thesis: Self-Supervised Learning and its Applications in Medical Image Analysis. Awarded doctoral degree with distinction.
M.Tech in VLSI & Microelectronics
Indian Institute of Engineering Science and Technology (IIEST), Shibpur, Howrah, India
Advisor: Dr. Ankita Pramanik
Dual Degree program with specialization in VLSI architectures and microelectronics.
B.Tech in Electronics & Telecommunication
Indian Institute of Engineering Science and Technology (IIEST), Shibpur, Howrah, India
Undergraduate curriculum covering signals, statistical systems, and communication engineering.
Impact & Global Network
Automated citation metrics synced with Google Scholar, coupled with an active collaborative network across France, Germany, Sweden, Hong Kong, and India.
Citations per Year
Auto-synced from Google Scholar
- Total Citations
- 246
- h-index
- 8
- i10-index
- 8
Loading chart…
International & National Collaborations
Partners in France, Germany, Sweden, Hong Kong, and leading Indian institutes
International
University of Lille 1 S&T, Lille, France
Dr. Tanmoy Mondal: Domain Adaptive Object Detection
International Research Collaborator
University of Hildesheim, Hildesheim, Germany
Dr. Koyel Ghosh: LLM Adaptation for Low-Resource Languages
International Research Collaborator
LUT University, Lappeenranta / Sweden Hub
Dr. Rajkumar Saini: AI in Fertility Science & Sperm Motility
International Research Collaborator
Hong Kong Baptist University, Kowloon, Hong Kong
Prof. Yiu-Ming Cheung: Federated Learning with Medical Segmentation
Post-Doc Collaboration / Former Institute
National
Indian Institute of Science (IISc), Bangalore, India
Prof. Anirban Chakraborty: PEFT-based FL Vulnerabilities & Inversion Attacks
Current Postdoctoral Host
Indian Statistical Institute (ISI), Kolkata, India
Prof. Umapada Pal: Self-Supervised Learning & Medical AI
Ph.D. Alma Mater & Ongoing Collaboration
IIT Kharagpur, Kharagpur, India
Dr. Saumik Bhattacharya: Medical Image Analysis & SSL Architectures
Academic Research Partner
NIT Durgapur, Durgapur, India
Dr. Sayantari Ghosh: Foreground Imbalance in Brachytherapy Segmentation
Academic Research Partner
Academic Service & Leadership
Contributing to the scientific community through various roles and initiatives
1. Editorial Activities
Journal Reviewer
- IEEE Transactions on Artificial Intelligence (TAI)
- IEEE Transactions on Multimedia (TMM)
- IEEE Access
- IEEE Internet of Things Journal (IoT)
- IEEE Transactions on Circuits and Systems for Video Technology (TCSVT)
- Elsevier Knowledge-Based Systems (KBS)
- Elsevier Pattern Recognition Letters (PRL)
- Elsevier Computer Vision and Image Understanding (CVIU)
- Springer Scientific Reports
- Springer Pattern Analysis and Applications
- Springer Multimedia Systems
- Springer International Journal of Machine Learning and Cybernetics
- Springer Neural Processing Letters
- Springer Nature Computer Science (SNCS)
- Transactions on Machine Learning Research (TMLR)
Conference Reviewer
- International Conference on Pattern Recognition (ICPR)
- ACM Knowledge Discovery and Data Mining (KDD)
- ACM Multimedia
- International Conference on Document Analysis and Recognition (ICDAR)
2. Teaching & Mentorship
Teaching Assistant
Introduction to Data Science (M. Tech. (CDS) 1st Year)
Indian Institute of Science, Bangalore,
- Probability
- Statistics
- Machine Learning
Artificial Intelligence and Machine Learning (M.Tech (CS))
Indian Statistical Institute, Kolkata,
- Object Detection
- Segmentation
- Machine Learning Concepts
- Project Evaluation on Deep Learning & CV
Introduction to Machine Learning (M.Tech (ETCE))
Indian Institute of Engineering Science and Technology, Shibpur,
- Machine Learning Algorithms
- Tutorial Sessions for First Semester Students
Mentorship
Ritik Kumar Badiya
IISc Bangalore, Ongoing
Gradient Inversion Attack
- Under Review
Ashmit Sinha
IISc Bangalore, Ongoing
Gradient Inversion Attack
- Under Review
Arpan Bairagi
ISI Kolkata, Ongoing
Heart Rate Monitoring
- ICIP 2026
Suyash Kumar
IIT BHU, Completed
Semi-Supervised Learning, Medical Imaging
- AI4Fertility Workshop
Priyangshu Mandal
IIT Kharagpur, Completed
Self-Supervised Learning
- Under Review
Soumitri Chattopadhyay
Jadavpur University, Completed
Signature Verification, Self-supervised Learning
- ICIP 2022
- ICPR 2022
- IEEE TAI 2024
Supreet Sahu
IIT Kharagpur, Completed
Self-Supervised Learning
- M.Tech. Dissertation
Sayan Das
ISI Kolkata, Completed
Self-supervised Medical Imaging
- M.Tech Dissertation
Tias Mondal
IIT Kharagpur, Completed
Self-Supervised Learning
- M.Tech. Dissertation
3. Tutorials & Invited Talks
Hands-On Tutorial Session
, MIDA 2023, SMIT, Sikkim & IDEAS-TIH, ISI Kolkata
Focused on self-supervised learning applications in medical imaging with practical implementations and case studies.
Curated Research Reading Lists
Carefully curated, active reading lists and literature repositories tracking the latest advances across key subfields. These static bibliographies serve as comprehensive entry points for researchers and students.
- Federated Machine Learning
Decentralized optimization, personalized client modeling, and non-IID data robustness.
- Inversion Attacks
Methods for reconstructing private client training inputs from model gradients and activations.
- Prompt Injection Attacks
Adversarial prompt injection vulnerabilities in vision-language models and LLMs.
- Machine Unlearning
Algorithmic data erasure and forgetting guarantees in trained deep models.
- Federated Unlearning
Selective revocation of client contributions without retraining global federated models.
- Unlearning in LLMs / VLMs
Concepts and benchmarks for targeted knowledge ablation in foundation models.
- Prompt Poisoning Attacks
Backdoor injections and malicious trigger mechanisms in prompt-tuned architectures.
- Vulnerability in Distributed Optimization
Byzantine fault tolerance, model poisoning, and convergence guarantees under attack.
AI Projects
Research Paper Summarizer
Active
Multi-modal document analysis leveraging VLM-based OCR and Gemma models for structured academic summarization.
- LLM
- VLM
- Document Understanding
- Scholarly Articles Discovery Interface
Active
System that aggregates newly published literature, filters topic relevance, and synthesizes annotated bibliographies.
- NLP
- Academic Search
- Bibliometrics
- Medical Image Viewer & Foundation Backend
In Progress
In-browser viewer supporting DICOM and NIfTI formats, integrated with foundation model backends for exploratory segmentation.
- Medical Image
- DICOM
- Segmentation
Blog: The Owl
- Updating your Local Branch from a Specific Remote Branch in Git
, 4 min read
Best practices for keeping local Git branches synchronized with upstream features cleanly.
- Running MedGemma-4B on a Small GPU (<16GB) Using BitsAndBytes
, 4 min read
Quantization strategies and memory budgeting for serving 4B parameter multimodal models locally.
- The Practical Guide to Distributed Training using PyTorch: Multi-Node with SLURM
, 6 min read
Configuring multi-node DDP workloads and master communication across academic cluster schedulers.