Dipankar Sarkar

💻

Dipankar Sarkar

Independent Researcher

About Me

Passionate researcher and technologist specializing in Machine Learning, Cryptography, and Distributed Systems. My work spans cutting-edge areas including federated learning, blockchain technology, and privacy-preserving AI systems. As the author of a book on Nginx (the world’s most popular web server), I bridge theoretical research with practical implementations. I combine academic rigor with industry experience, having architected large-scale systems and led technical innovations at major technology companies. My current research focuses on decentralized AI systems, privacy-preserving machine learning, and the development of robust distributed infrastructure.

Download CV

Interests

Machine Learning
Cryptography
Distributed Systems
Federated Learning
Augmented Reality
Virtual Reality
Image Processing

Education

MSc Computer Science
Arizona State University
Graduate Certificate, Strategic Studies
The Takshashila Institute
BTech Computer Science
Indian Institute of Technology

📚 My Research

My research lies at the intersection of machine learning, distributed systems, and decentralized technologies. As a computer scientist and entrepreneur, I focus on developing practical solutions that bridge theoretical advances with real-world applications.

In the realm of federated learning, I’ve pioneered techniques for handling imbalanced data classification and optimizing communication efficiency. My work on Fed-Focal Loss and CatFedAvg has contributed to making distributed machine learning more practical and efficient. My recent work extends into Web3 and decentralized systems, where I’ve developed protocols for physical infrastructure networks (DePIN) and mechanisms for fair value distribution in blockchain environments. I’ve published several papers on atomic composability in multi-rollup environments and designed novel approaches for MEV mitigation in Ethereum.

I actively collaborate on projects involving federated learning, decentralized systems, and privacy-preserving AI, continuously working to advance the field while maintaining a focus on practical applicability.

Featured Publications

Large Language Models

Navigating the Knowledge Sea: Planet-scale answer retrieval using LLMs

This research investigates the application of Large Language Models (LLMs) in developing scalable solutions for information retrieval across extensive knowledge bases. The work addresses fundamental challenges in managing and retrieving information at a planetary scale, combining the power of modern language models with efficient retrieval mechanisms.

Feb 1, 2024

Generative AI

Viz: A QLoRA-based Copyright Marketplace for Legally Compliant Generative AI

This paper introduces Viz, a novel marketplace framework that leverages QLoRA (Quantized Low-Rank Adaptation) to enable copyright-compliant generative AI systems. We present a technical solution for tracking and managing intellectual property rights in AI-generated content while maintaining model efficiency through quantization.

Dec 1, 2023

Blockchain

Decentralized Deepfake Detection Blockchain Network using Dynamic Algorithm management

Abstract This paper presents a novel approach to deepfake detection through a decentralized blockchain network that dynamically manages detection algorithms. The system leverages blockchain technology to create a transparent and secure platform where multiple deepfake detection algorithms can be deployed, evaluated, and updated in real-time, ensuring adaptability to new deepfake techniques.

Nov 1, 2023

DePIN

Generalised DePIN Protocol: A Framework for Decentralized Physical Infrastructure Networks

The Generalised DePIN Protocol paper introduces a comprehensive framework for Decentralized Physical Infrastructure Networks (DePIN). This research presents a theoretical foundation for understanding and implementing DePIN systems, which bridge the gap between physical infrastructure and blockchain technology.

Nov 1, 2023

Federated Learning

Fed-Focal Loss for imbalanced data classification in Federated Learning

Federated Learning has emerged as a promising paradigm for training machine learning models while preserving data privacy. However, handling class imbalance in federated settings remains challenging. This work introduces Fed-Focal Loss, a novel approach that adapts focal loss for federated learning scenarios to address data imbalance across distributed clients.

Nov 1, 2020

See all publications

Patents

Virtual Reality

A Method and a System for Dynamically Modifying a Virtual World

Virtual worlds have become increasingly important in applications ranging from gaming and entertainment to training and simulation. However, traditional virtual environments often suffer from static nature and limited adaptability. This invention addresses these limitations by introducing a novel method and system for dynamically modifying virtual worlds in real-time.

Aug 7, 2020

Virtual Reality

A Method and System to Recommend a Navigation Path in a Virtual World

This patent addresses a fundamental challenge in virtual reality environments: helping users navigate efficiently through complex virtual worlds. As virtual environments become increasingly sophisticated and expansive, users often struggle to find optimal paths to their destinations or discover interesting locations within these digital spaces.

Aug 7, 2020

Virtual Reality

A System to Generate and Dynamically Update a Virtual World Based on User Interest

This patent introduces an innovative system that revolutionizes how virtual worlds adapt to individual users. The core idea is to create virtual environments that evolve and transform based on real-time user engagement and interests, rather than remaining static experiences.

Aug 7, 2020

Animation

Method and System to Generate Animated Audio-Visual Content

This patent introduces an innovative method for automatically generating animated audio-visual content, addressing a growing need in the digital content creation space. As the creator of this technology, I aimed to streamline the labor-intensive process of creating animated content while maintaining high quality and engagement.

Jul 10, 2020

Computer Graphics

Method for Converting Static Graphic into Animated Graphic

This patent presents an innovative method for automatically transforming static images into engaging animated graphics, addressing a significant challenge in digital content creation. As the demand for dynamic visual content continues to grow across social media, marketing, and educational platforms, the manual process of creating animations remains time-consuming and requires specialized skills.

Jul 10, 2020

See all

Recent & Upcoming Talks

Federated Learning

Fed-Focal Loss for Imbalanced Data Classification in Federated Learning

A presentation on applying Focal Loss to Federated Learning for handling imbalanced data classification

Jan 8, 2021

Fragaria: From 'r's in Strawberry to Complex Problem-Solving AI

Mar 19, 2024

Introducing Fragaria, an advanced Chain of Thought (CoT) reasoning API that combines LLM capabilities with Reinforcement Learning to tackle complex problem-solving challenges.

Mar 19, 2024