[CVPR 2024 Award Candidate] Producing and Leveraging Online Map Uncertainty in Trajectory Prediction

[ECCV 2024] Accelerating Online Mapping and Behavior Prediction via Direct BEV Feature Attention

map estimation of topic models

DRAM error-correction code (ECC) simulator incorporating statistical error properties and DRAM design characteristics for inferring pre-correction error characteristics using only the post-correction errors. Described in the 2019 DSN paper by Patel et al.: https://people.inf.ethz.ch/omutlu/pub/understanding-and-modeling-in-DRAM-ECC_dsn19.pdf.

Implementation of framework and reproduction of figures from "A Modularized Efficient Framework for Non-Markov Time Series Estimation" (https://arxiv.org/abs/1706.04685)

This is a repository with the assignments of IE675b Machine Learning course at University of Mannheim.

Assignment 5, Data Analysis and Interpretation, Autumn 2020, IIT Bombay

Assignments completed for my Machine Learning course: Topics include probability and statistics proofs, MLE/MAP parameter estimation, EM Algorithm, Bayes Theorem implementations, gradient descent methods, Neural Networks and Deep Learning.

Binary classification project in Python comparing a MAP classifier (with known distributions) vs. logistic regression. Includes analytical posterior derivation, Monte Carlo simulation, and gradient-based training. Focus on error probability, variance impact, and model performance.

Python library and TRIPS-Py examples for prior-normalizing (Knothe-Rosenblatt) transport maps in sparse Bayesian learning (SBL): MAP demos for 1D/2D deblurring alongside Tikhonov/GCV baselines. Companion to Glaubitz & Marzouk (2025) and the Julia reference implementation.

2-D robot localization using SE(2) factor graphs and robust MAP estimation.

Bayesian logistic regression for predicting NYC CitiBike trip destinations, comparing global and station-specific models on 5.2M trips with MAP estimation via PyMC.

To alleviate unbounded drift in model updates, we proposed a "Bayesian fine-tuning using a reference model"(FedRef)