CV
Contact Information
| Name | Satya Prakash Dash |
| Professional Title | PhD Researcher in Computer Science |
| sprakashdash@gmail.com | |
| Phone | +44 7423727177 |
Professional Summary
PhD researcher at University of Manchester specializing in second-order optimization, large language models, and deep reinforcement learning. Passionate about advancing machine learning through theoretical understanding and practical applications.
Education
-
2023 - 2026 Manchester, UK
PhD
The University of Manchester
Computer Science
- Supervised by Prof Sami Kaski, Mingfei Sun & Wei Pan
- Fully funded via Dean’s Doctoral Scholarship
- Research focus: Second-order optimizer trajectories using Fisher Information Matrix for fast LLM adaptation
- Recent paper submitted to AISTATS 2026 on Gradient Regularized Natural Gradient
-
2015 - 2020 Kharagpur, India
B.Sc. & M.Sc. (Hons) First Class
Indian Institute of Technology, Kharagpur
Physics
- Received Inspire Scholarship (2015-2020) by DST, Government of India
- Strong foundation in theoretical physics and computational methods
-
2013 - 2014 India
Senior Secondary
Central Board of Secondary Education, India
Computer Science Stream
- Selected for INMO; cleared Regional Mathematical Olympiad 2014
Experience
-
2022 - 2022 Remote
Sr. Machine Learning Research Engineer
Visionify.AI
Developed and deployed machine learning solutions for retail computer vision applications
- Integrated wandb.ai into classification and object detection model training pipelines
- Built and deployed perspective correction algorithm for real-time footage
- Built blur detection pipeline to enhance out-of-stock accuracy
- Managed quality control of deep image models
-
2021 - 2022 India
Machine Learning Research Engineer
Expand-AI Pvt Ltd
Established few-shot learning pipeline for efficient data annotation
- Established pipeline achieving 70% faster turnaround than manual annotation
- Onboarded three clients and produced over million annotations
- Managed and assessed team of ten data associates
-
2025 - 2025 Remote
LLM for Nutritional Science
Brain-Feed Pvt. Ltd.
Developed LLM solutions for nutritional science research
- Built Prompt-Tuning and RAG pipeline using AWS BedRock
- Developed pipeline for in-depth metadata analysis
- Implemented fine-tuning methods for adapting Claude LLM
-
2022 - 2022 Cyprus
Advanced Quadrupedal Locomotion from Vision through Deep RL
CYENS, Cyprus
Developed vision-based reinforcement learning for quadruped robots
- Worked with IsaacGym simulator for robot locomotion
- Built vision module (CNN + LSTM) for obstacle avoidance
- Designed complex terrains to test controller robustness
-
2018 - 2019 Kharagpur, India
Spatial Variability of Ammonia & Particulate Matter Hotspots in India
IIT Kharagpur
Analyzed atmospheric pollution data using satellite imagery
- Collected decade of Ammonia data from Metop-A satellite
- Built preprocessing and visualization pipelines
- Research contributed to Science of Total Environment publication
Awards
-
2023 Dean's Doctoral Scholarship
University of Manchester
Full PhD funding award for the complete term of doctorate
-
2015 Inspire Scholarship
DST, Government of India
Merit-based scholarship awarded through JEE Advanced 2015
-
2016 Social & Cultural Secretary
Azad Hall of Residence, IIT Kharagpur
Elected to lead social and cultural activities
-
2014 Regional Mathematical Olympiad
Mathematical Olympiad Selection
Cleared RMO 2014 and selected for INMO
Publications
-
2025 Rank-1 Approximation of Inverse Fisher for Natural Policy Gradients in Deep Reinforcement Learning
TMLR 2025
Accepted paper on natural policy gradients in deep reinforcement learning
-
2026 Gradient Regularized Natural Gradients
AISTATS 2026
Accepted to AISTATS 2026 on scalable natural gradients using gradient regularization
-
2025 Guided Riemannian Optimization (GuRO): Bridging Model Predictive Control and Decision Transformers
ICRA 2025
Submitted to ICRA 2025 on control and decision transformers
-
2020 Record-high levels of Atmospheric Ammonia over India: Spatial and Temporal Analysis
Science of Total Environment, Elsevier
Published research on atmospheric ammonia levels over India
Skills
Programming Languages (Expert): Python (PyTorch, Keras, TensorFlow2, OpenCV, Numpy, Scikit-Learn, Flask), C++ (STL, CUDA, OpenMP)
Tools & Frameworks (Advanced): ROS, MuJoCo, PyBullet, VS Code, JupyterLab, GCP, AWS-BedRock, Docker, Git, Matlab, LaTeX
Research Areas (Advanced): Machine Learning, Deep Reinforcement Learning, Natural Gradient Descent, Large Language Models, Computer Vision, Optimization Theory
Languages
English : Fluent
Hindi : Fluent
Certificates
- CUDA Certification - NVIDIA (2024)
- Cambridge Ellis Unit Summer School - University of Cambridge (2023)
- NeuroMatch Academy Deep Learning Course - NeuroMatch Academy (2021)
Projects
-
Fixed-Fisher for Pre-Training LLMs
Selected for Deepmind - RAEng Research Ready Summer internship. Proposes scalable Natural Gradient Descent for faster LLM pre-training
- Selected for Royal Academy of Engineering Research Ready Summer internship
- Implemented scalable Natural Gradient Descent for pre-training GPT2 and ViT
- Achieved superior performance compared to AdamW training
-
Policy Gradient Algorithms in PyTorch - RL.Fun.Do
Comprehensive implementation of reinforcement learning algorithms with theoretical framework based on Statistical Physics
- Implemented DDPG, PPO, TRPO and SAC for OpenAI Gym and TORCS
- Formulated theoretical framework connecting RL to Statistical Physics using partition functions
- Masters thesis: Continuous Control in Deep Reinforcement Learning and Statistical Physics
-
Computational Neuroscience
Simulated auditory and neural dynamics using computational models
- Simulated rate response and tuning curves of Auditory Nerve Fibre using Zilany et al. model
- Examined non-linear dynamical spiking through Morris-Lecar equations and Hodgkin-Huxley Model