Abstract: AI Researcher and Builder at Lossfunk, founded: HyperSym

Hi, I am Siddhant Saxena (@Sidgraph), currently a Pre-Doctoral Researcher/ Builder in Agentic systems, Applied LLMs, Reasoning Models, Agentic Graph Systems, Geometric Deep Learning (GDL), RL, and Large Parametric Models (all modalities-mainly geometry).

I am also affiliated with VISTA Lab, Indian Institute of Sciences, Banglore (IISC-BLR), working on Learning on large graphs, Noisy-data distributional learning, core-problems in GraphML.

I was an Applied Scientist at The Agentic, working on Function Calling, Foundational Models (Geometry), LLM Inference, and Chip designs.

Previously I was an Applied Scientist at Snapdeal, Leading large-scale industrial projects like: Recommendation Engine, RTO Modelling.

I recently graduated as a Pre-Doctoral researcher at LCS2 Lab in the field of Geometric Deep Learning, advised by Dr. Tanmoy Charkaborty and Dr. Subhabrata Dutta from India.
{Category Theory, Differential Geometry, Algebraic Topology, Modern Algebra, Algebraic Combinatorics}

visit Activities for Events and Learnings
I am *not* applying to Ph.D. admissions this fall [2024], I will be working as an associate researcher [pre-doc] at Lab. Currently working on Riemannian Manifold Learning, Geometric DL, and Category Theory related projects.

Coming from a Third World Country, I have to do Freelance and Consulting as well. I enjoy working in a wide range of projects, "All things in AI"

Recent

New Paper: DPHGNN: A Dual Perspective Hypergraph Neural Networks [KDD'24]
Working on Representation Learning on Higher Geometrical Structures [META AI Research Funded]
Working on Few-Shot Graph Incremental Learning in Graphs [META AI Research Funded]

Learning Category Theory and Functional Analysis on Riemannian Manifolds

A Paper A Day keeps Forgetting Away

Twitter

Blog

Schedule a Meet :)

Annotated Papers

About me

Hi, I am Siddhant Saxena currently an undergraduate researcher in the field of Computational Intelligence from India.

Note: All the works I have done are from my solo contribution, I thank my advisors for their constant support and collaborations.

Current Research Interests:

Geometric Deep Learning
Reinforcement Learning
Molecular Dynamics
Physics-inspired Deep Learning
Category Theory
Differential Algebraic Geometry
Neural Algorithmic Reasoning
Hypergraph Representation Learning
Contextual - Geometric Reasoning in LLMs
3D Reconstruction Problems

Current Affiliations

Meet Simmy, your AI-Powered Simulation Co‑Pilot 🚀

At Lossfunk, we're building HyperSym, Agents accelerating the entire design, simulation, and validation lifecycle for Hard sciences.

Activities Timeline

I am regularly updating the portfolio and hence I post only recent news here, kindly visit the activities page to know what are the exciting things I do in a month-wise timeline fashion :)

Activities Timeline

Research Collaborations

[Pre-Doctoral Researcher] LCS2-Lab: IIT Delhi

[Pre-Doctoral Researcher] VISTA Lab: IISC-BLR

Applied Scientist [The Agentic]

[Summer Researcher] ACM IKDD Special Interest Group

Agentic Graph Systems
Knowledge Graph based Meta RL agents that perform function calling in more cost and reasoning efficient approaches.
Test-Time Inference Optimizations
Chip Design

Temporal Graph Recommendation Systems
Establishment of biases in temporal graph amplification loop
Graph Variational Auto Encoders based debiaising architecture.
Learning to de-bias in temporally evolving graphs

[Winter Researcher] Trust Lab IIT Bombay

Applied Scientist

Neural Cryptographic Reasoning
Exploring Neural Algorithmic Reasoning for Cryptographic Algorithms
Developed NCR - Encryption - Decryption Algorithms

TabularGraphTransformer for RTO
Graph-Transformer employed RTO prediction task for million-size graphs.
Graph Neural Networks for Recommendation Systems

Research Publications and Research Projects

[Graph Machine Learning, OSM-OSQ Dillema, Riemannian Geometry]

Task-aware Latent Graph Rewiring Can Robustly Solve Oversquashing-Oversmoothing Dilemma. [Pre-Print]
Lead contributor. End-to-End work on solving core problems in Graph Machine Learning, developed SOTA architecture to robustly solve over-smoothing and over-squashing problems concurrently.

In collaboration with Meta AI (SF team)

[Graph Machine Learning, Hypergraph Neural Networks, Return-to-Origin]

DPHGNN: A Dual Perspective Hypergraph Neural Networks [KDD'24]
Lead contributor. Developed full architecture, performed extensive experiments, developed theoretical frameworks, and wrote the full paper.

In collaboration with Meesho

[Graph Machine Learning, Recommendation Systems, Statistics]

Biases and De-Biasing Strategies for Temporally evolving Graph RecSys

Contribution: developed temporally evolving graph recsys, establish and eliminate exposure bias in the recommendation policy.

Used Pytorch and Pytorch Geometric for Implementation

[Graph Machine Learning, Network Science]

Understanding Evolution through Spatio-Temporal Graph Generation

Contribution: Developed a robust unsupervised graph generation framework to understand the evolution of social networks.

Used Pytorch and Pytorch Geometric for Implementation

[Generative Deep Learning, Network Security, IOT]

ImmuneGAN: A self-adaptive artiﬁcial immune system for network intrusion detection.

Selected for Publication: International Conference on Cyber Security, Privacy and Networking (ICSPN2022), Thailand

Contribution: Application of GAN-based architecture in data security and implemented ImmuneGAN in PyTorch for tabular data

[Timeseries forecasting. Deep Learning, Smart City]

GRUBin: Time-series forecasting based on eﬃcient garbage monitoring for smart city ecosystem

Submitted in International Journal of Integrated Waste Management, Science and Technology

Contribution: Implemented Time-series forecasting module to predict timestamps of garbage fill-up in PyTorch

[Deep Learning, IoT, Network Security]

DAIS: Deep Artiﬁcial Immune System for Intrusion Detection in IoT Network Ecosystem

Selected for Publication: International Journal of Bio-inspired Computing

Contribution: Development of Neural networks based on the artificial immune system and statistical analysis of network packets, implemented in PyTorch

[5] A Mixed Deep Learning and Statistical Approach for Network Anomaly Detection.

Book Chapter: Routledge Tayler and Francis Group

Contribution: Book chapter writing on Network Anomaly Detection Approaches

[6] A Review on Security and Privacy Requirements and Attacks in VANET

Paper Presentation: International Conference on Cyber Warfare, Security & Space Research (SpacSec-2021)

Contribution: Reviewed the attacks on VANETs and Privacy Requirements

Independent Collaborations and Consultancy

[Deep Learning, IoT, Network Security]

DAIS: Deep Artiﬁcial Immune System for Intrusion Detection in IoT Network Ecosystem

Selected for Publication: International Journal of Bio-inspired Computing

Contribution: Development of Neural networks based on the artificial immune system and statistical analysis of network packets, implemented in PyTorch

[Timeseries forecasting. Deep Learning, Smart City]

GRUBin: Time-series forecasting based on eﬃcient garbage monitoring for smart city ecosystem

Submitted in International Journal of Integrated Waste Management, Science and Technology

Contribution: Implemented Time-series forecasting module to predict timestamps of garbage fill-up in PyTorch

Genesis AI Labs

I also lead an undergraduate research group, producing cutting-edge research in various areas of Computational Intelligence. We are a group of researchers looking forward to establishing a distinguished research lab in India, to encourage and support early carrier researchers in terms of mentorship, computing resources, collaboration, and resources. As the undergrad and masters students seek good collaborators and mentors, we have developed research pods to encourage interdisciplinary research in artificial intelligence and we are collaborating with well-established research labs and researchers to provide a cutting-edge research environment to our researchers.

At Genesis AI Labs we work on emerging technologies in the field of artificial intelligence, and participate and organize various events to learn and collaborate with people working in other domains, we also organize learning groups and paper discussion sessions to learn and grow together.

We believe that learning and innovation depend on the orientation of young minds, thus we are open to students from any university across the globe based on the knowledge representation by individual and not on status.

Google Sites

Report abuse