Causal Analysis of Corner Kicks
A causal framework for the analysis of Corner Kicks, using the combination of Opta tracking & events data.
Interested in projects that are at the intersection of Data & Machine Learning.
I am currently working as an AI & Machine Learning Engineer at Aston Villa Football Club.
I hold a Masters degree in Computer Science alongside a comprehensive skill set acquired during my professional experiences at Aston Villa, KnetMiner and Hewlett Packard Enterprise. My professional journey has equipped me with valuable skills in automating infrastructure, utilizing RESTful APIs, and deploying in a CI/CD environment.
Additionally, my academic pursuits have focused on applying Machine Learning algorithms to Vision and NLP datasets. My Master's Thesis on training Score-based generative models with Non-Gaussian noise can be found on Indigo-UIC. The research was also selected for poster presentation at the MMLS 2023 conference. You can find the pre-print version of this paper on arxiv .
If you want to know more about me or my work, please don't hesitate in reaching out via, email: mishraharsh169@gmail.com
ResumeAchievements
- A Causal Analysis of Corner Kicks, 2024, OPTA FORUM - Harsh Mishra*
- Flag Aggregator: Scalable Distributed Training under Failures and Augmented Losses using Convex Optimization, 2024, ICLR - Hamidreza Almasi*, Harsh Mishra, Balajee Vamanan, Sathya N. Ravi
- Using Intermediate Forward Iterates for Intermediate Generator Optimization, 2023, MMLS - Harsh Mishra*, Jurijs Nazarovs, Manmohan Dogra, Sathya N. Ravi
Projects
Detection and analysis of corner situations using only broadcast tracking data.
Used standard data pre-processing and cleaning techniques, applied K-means to cluster player locations into zones.
Detect corner situations in soccer tracking data
A causal analysis pipeline to identify factors that influenced public sentiment during the COVID-19 pandemic.
We present a comparison between NoTears, NoCurl and NoFear algorithms to find causal graphs and then use Bayesian Networks to find the conditional probabilities.
Causal Inference on Reddit Data
Training Score-based generative models with Non-Gaussian noise.
We present a novel architecture and show its utility in image generation, point cloud denoising and generative PCA.
Generative Score based models using non-Gaussian noise
End to end Log Analysis Pipeline.
AWS EBS - generate logs, Hadoop Spark & MapReduce - data crunching, Kafka - real time streaming of logs, D3.js - real time visualization.
Log Streaming Pipeline
Interactive visualization of protein sequence data - D3.js
Used SDLC and data cleaning & visualization techniques to come up with ways to visualize amino acid protein data.
Visualization of Amino acid protein structure
Algorithm to convert categorical labels/features to continuous labels.
The algorithm allows the use of kernel methods for node classification and other GNN tasks. Applications also on Event Stream and Pose Estimation data.
Algorithm for Information propagation and node labeling in Graph Neural Networks
End to end tweet extraction and summarization.
Extracts tweets for a given keyword, in a given time period, and produces a concise summary along with the tweet sentiments. Model available on hugging face.
Twitter Opinion Summarization
Find similar soccer players using UMAP & GMM.
Applied UMAP to convert soccer events data to smaller dimension, then compared GMM's clusters to actual player positions.
Soccer Player Position Analysis
Open source contribution.
Made the original code compatible with any given dataset, fixed versioning issues, and changed the output format required by "Controllable Video Generation with Sparse Trajectories", CVPR'18.
Video frame generation using Optical Flow
Get In Touch
If you want to know more about me or my work, please don't hesitate in reaching out.