# Dipankar Sarkar — Independent Researcher

> Researcher in machine learning and decentralized systems. Federated learning, privacy-preserving AI,
> blockchain infrastructure, MEV mitigation, DePIN protocols. ACM and IEEE member with 138+ citations.

## Identity & Disambiguation
Full name: Dipankar Sarkar. IIT Delhi (B.Tech Computer Science, 2003-2007) and Arizona State University (M.S. Computer Science, Cybersecurity specialization, 2020-2022) alumnus.
Author of the Fed-Focal Loss paper, the Nginx Web Server Implementation Cookbook (Packt Publishing, 2011), and 25+ provisional patent applications filed with the Indian Patent Office.
Not to be confused with the Indian film critic or cricketer of the same name.

## Research Focus
- Federated Learning (Fed-Focal Loss, CatFedAvg communication efficiency)
- Privacy-preserving distributed machine learning
- Decentralized physical infrastructure networks (DePIN)
- MEV mitigation in Ethereum (FairFlow Protocol)
- Multi-rollup composability and cross-chain protocols
- Cryptography and zero-knowledge proofs

## Other Sites by Dipankar Sarkar
This is the academic / research persona site. Dipankar maintains four sister sites for distinct facets of his work:
- [dipankar.name](https://www.dipankar.name) — engineering leader, AI/ML builder, Rust open source (canonical hub)
- [dipankar.org](https://www.dipankar.org) — startup consultant, angel investor, venture case studies
- [dipankar.co](https://www.dipankar.co) — fractional CTO, technology consulting, client case studies
- [desinerd.com](https://www.desinerd.com) — personal blog since 2007, AI/DevOps/self-hosting

## Publications
- Test-Input Generation for Tensor Programs: What Actually Finds Kernel Bugs (2026-06-20) — arXiv preprint arXiv:2606.27396. https://www.dipankar.cc/publication/test-input-generation-tensor-programs.md/
- The Correctness Illusion in LLM-Generated GPU Kernels (2026-06-15) — arXiv preprint arXiv:2606.20128. https://www.dipankar.cc/publication/correctness-illusion-llm-gpu-kernels.md/
- Before the Pull Request: Mining Multi-Agent Coordination (2026-06-10) — arXiv preprint arXiv:2606.19616. https://www.dipankar.cc/publication/before-the-pull-request-multi-agent-coordination.md/
- An Input-Regime Audit of Conflict Detection for Retrieval-Augmented Generation (2026-05-15) — arXiv preprint (submitted to VecDB@VLDB 2026). https://www.dipankar.cc/publication/input-regime-audit-rag-conflict-detection.md/
- Epistral Network: Revolutionizing Media Curation and Consumption through Decentralization (2024-02-01) — arXiv preprint arXiv:2402.04881. https://www.dipankar.cc/publication/epistral-network.md/
- Navigating the Knowledge Sea: Planet-scale answer retrieval using LLMs (2024-02-01) — arXiv preprint arXiv:2402.05318. https://www.dipankar.cc/publication/navigating-knowledge-sea.md/
- FairFlow Protocol: Equitable Maximal Extractable Value (MEV) mitigation in Ethereum (2023-12-01) — arXiv preprint arXiv:2312.12654. https://www.dipankar.cc/publication/fairflow-protocol.md/
- Viz: A QLoRA-based Copyright Marketplace for Legally Compliant Generative AI (2023-12-01) — arXiv preprint arXiv:2401.00503. https://www.dipankar.cc/publication/viz-qlora-copyright-marketplace.md/
- Centralized Intermediation in a Decentralized Web3 Economy: Value Accrual and Extraction (2023-11-01) — arXiv preprint arXiv:2311.08234. https://www.dipankar.cc/publication/centralized-intermediation-web3.md/
- Decentralized Deepfake Detection Blockchain Network using Dynamic Algorithm management (2023-11-01) — arXiv preprint arXiv:2311.18545. https://www.dipankar.cc/publication/decentralized-deepfake-detection.md/
- Generalised DePIN Protocol: A Framework for Decentralized Physical Infrastructure Networks (2023-11-01) — arXiv preprint arXiv:2311.00551. https://www.dipankar.cc/publication/generalised-depin-protocol.md/
- Towards Universal Atomic Composability: A Formal Model for Multi-Rollup Environments on Ethereum (2023-11-01) — arXiv preprint arXiv:2311.00422. https://www.dipankar.cc/publication/universal-atomic-composability.md/
- Curriculum generation using Autoencoder based continuous optimization (2021-06-01) — arXiv preprint arXiv:2106.08569. https://www.dipankar.cc/publication/curriculum-generation-autoencoder.md/
- One Shot Audio to Animated Video Generation (2021-02-01) — arXiv preprint arXiv:2102.09737. https://www.dipankar.cc/publication/one-shot-audio-animation.md/
- Catfedavg: Optimising communication-efficiency and classification accuracy in federated learning (2020-11-01) — arXiv preprint arXiv:2011.07229. https://www.dipankar.cc/publication/catfedavg.md/
- Fed-Focal Loss for imbalanced data classification in Federated Learning (2020-11-01) — Workshop on Federated Learning for Data Privacy and Confidentiality in Conjunction with IJCAI 2020. https://www.dipankar.cc/publication/fed-focal-loss.md/
- Nginx 1 Web Server Implementation Cookbook — Packt Publishing (2011) (2011-01-01) — Packt Publishing. https://www.dipankar.cc/publication/nginx-cookbook.md/
- Muzzled: A graphical theme builder for Mozilla (2005-01-01) — Linux Gazette. https://www.dipankar.cc/publication/muzzled-mozilla-theme-builder.md/

## Patents (Provisional Applications, Indian Patent Office)
- A Method and a System for Dynamically Modifying a Virtual World — Provisional Application (Indian Patent Office, 2020-08-07)
- A System to Generate and Dynamically Update a Virtual World Based on User Interest — Provisional Application (Indian Patent Office, 2020-08-07)
- A Method and System to Recommend a Navigation Path in a Virtual World — Provisional Application (Indian Patent Office, 2020-08-07)
- Method and System to Generate Animated Audio-Visual Content — Provisional Application (Indian Patent Office, 2020-07-10)
- System and Method for Partitioning a Neural Network Model for Offloading Computational Load — Provisional Application (Indian Patent Office, 2020-07-10)
- Method for Converting Static Graphic into Animated Graphic — Provisional Application (Indian Patent Office, 2020-07-10)
- Method for Dynamic Content Generation and Device Thereof — Provisional Application (Indian Patent Office, 2020-06-30)
- A Method and System for Partitioning a Social Network Group — Provisional Application (Indian Patent Office, 2020-05-06)
- A Method and System for Improvising a Pre-Generated Avatar of a User — Provisional Application (Indian Patent Office, 2020-04-24)
- A System to Generate an In-Context Link to a Shared Experience in a Chat Session — Provisional Application (Indian Patent Office, 2020-04-22)
- System to Generate and Recommend Offline Compatible Interactions to Users — Provisional Application (Indian Patent Office, 2020-04-21)
- Methods and Systems for Impersonation and Exploration of a Virtual World — Provisional Application (Indian Patent Office, 2020-04-21)
- A System to Alter User Profile and Experience on a Virtual Communication Platform — Provisional Application (Indian Patent Office, 2020-04-21)
- Method of Intimating Experience/Activity to a User — Provisional Application (Indian Patent Office, 2020-04-17)
- Method and System for Transmitting Messages in Low Bandwidth Environment — Provisional Application (Indian Patent Office, 2020-04-14)
- A System and Method for Generating a Video/Animation for Matched Users — Provisional Application (Indian Patent Office, 2020-04-14)
- Methods and Systems to Transform a User Recorded Video into a Video Sticker — Provisional Application (Indian Patent Office, 2020-02-24)
- System and Method for Generating Animated Visual Appearance of User, Based on Audio Message — Provisional Application (Indian Patent Office, 2020-02-18)
- A System and Method for Generating Unified Image on a Messaging Platform — Provisional Application (Indian Patent Office, 2020-02-18)
- Systems and Methods for Converting Text to Speech Mimicking User's Voice Tone — Provisional Application (Indian Patent Office, 2020-02-03)
- A Method and System for Generating Multiple Expressive Emojis — Provisional Application (Indian Patent Office, 2020-01-06)
- A System for Providing Avatars from an Encrypted Image of a User — Provisional Application (Indian Patent Office, 2020-01-06)
- A Method and System for Generating Hairstyle Vector — Provisional Application (Indian Patent Office, 2020-01-06)
- Natural Language Query Based Search System — Provisional Application (Indian Patent Office, 2020-01-06)
- A Method and a System for Determining a Relationship Type Between Users — Provisional Application (Indian Patent Office, 2020-01-06)

## Posts
- [Automated Prompt Optimization: From AutoPrompt (2020) to TextGrad (2024)](https://www.dipankar.cc/post/automated-prompt-optimization-survey.md/) — A chronological survey of automated prompt optimization 2020–2025: AutoPrompt, APE, OPRO, EvoPrompt, DSPy, TextGrad, PromptAgent, and how to choose between them
- [LLM Prompt Compression: LLMLingua, GIST Tokens, and the Path to 480x Compression](https://www.dipankar.cc/post/llm-prompt-compression-guide.md/) — A practitioner's guide to LLM prompt compression: LLMLingua, GIST Tokens, 500xCompressor, KV-cache methods, and the rate-distortion limits of compressing contex
- [Prompt Structuring Techniques: From Chain-of-Thought to the Instruction Hierarchy](https://www.dipankar.cc/post/llm-prompt-structuring-techniques.md/) — A chronological survey of LLM prompt structuring: chain-of-thought, the instruction hierarchy, system prompt design, evaluation frameworks, and the theoretical 
- [LLM Safety Techniques: Constitutional AI, Harmony, SAIF, and Llama Guard Compared](https://www.dipankar.cc/post/llm-safety-techniques-survey.md/) — A practitioner's survey of LLM safety techniques across OpenAI Harmony, Anthropic Constitutional AI, Google SAIF, Meta Llama Guard, and open-source RLHF framewo
- [Tackling Data Imbalance in Federated Learning](https://www.dipankar.cc/post/federated-learning-imbalanced-data.md/) — How Fed-Focal Loss addresses one of the most challenging problems in distributed machine learning: handling imbalanced data across federated clients.
- [The AI Copyright Challenge: Building Legal Frameworks for Generative AI](https://www.dipankar.cc/post/ai-copyright-challenge.md/) — As generative AI transforms content creation, we need new frameworks that respect copyright while enabling innovation. Here's how we can build them.
- [The MEV Problem: Why Ethereum Needs Fairer Value Distribution](https://www.dipankar.cc/post/mev-ethereum-fairness.md/) — Exploring Maximal Extractable Value (MEV) in Ethereum and why we need better mechanisms for fair value distribution across the ecosystem.
- [DePIN: The Future of Physical Infrastructure](https://www.dipankar.cc/post/depin-future-infrastructure.md/) — Why Decentralized Physical Infrastructure Networks (DePIN) represent a fundamental shift in how we build and own critical infrastructure.
- [Why Deepfake Detection Needs Decentralization](https://www.dipankar.cc/post/deepfake-detection-decentralized.md/) — As deepfake technology becomes more sophisticated, centralized detection approaches are failing. Here's why we need decentralized solutions.

## Contact
- Website: https://www.dipankar.cc
- Email: contact@dipankar.cc
- ORCID: https://orcid.org/0000-0001-5431-6367
- Google Scholar: https://scholar.google.com/citations?user=t_ikr2UAAAAJ&hl=en
- GitHub: https://github.com/sarkar-dipankar (research) / https://github.com/dipankar (Rust CLIs)
- LinkedIn: https://www.linkedin.com/in/dipankarsarkar