A text diffusion model based on DDPM operating in embedding space, generating coherent sequences by denoising continuous latent representations instead of discrete tokens.
-
Updated
May 5, 2026 - Python
A text diffusion model based on DDPM operating in embedding space, generating coherent sequences by denoising continuous latent representations instead of discrete tokens.
Code for ASD-FEAT: a multi-modal, video-derived infant dataset for early ASD risk prediction. Two-stage pipeline — transformer behavior detection (look-face, look-object, smile, vocalization) followed by risk classification on per-session aggregate features.
Deep Reinforcement Learning (PPO) for autonomous driving on multilane highways using HighwayEnv, comparing safety-regularized reward shaping against a deterministic heuristic benchmark.
An unforced, asynchronous 3D Navier-Stokes & Euler operator engine implementing runtime verification of finite-time blow-up attractors in Sobolev spaces, bypasses Python's lack of Tail-Call Optimization (TCO), eliminates heap allocation churn, and delivers deterministic O(1) space complexity for infinite data streams and complex PDEs.
Guarded, tool-using banking-support agent for a fictional bank: multi-turn function calling, RAG over Regulation E, prompt-injection/PII/action-policy guardrails, an eval harness with red-teaming, LoRA fine-tuning and vLLM inference benchmarks. Synthetic data only.
Computer vision system for tyre defect detection — MobileNetV2, 89.6% accuracy, Streamlit dashboard
PPO-GNN framework for stochastic fuel delivery optimization. Achieves 1.9% optimality gap with 40× speedup vs exact solvers.
some practise codes on machine learning and courses assignments
This project explores multimodal AI for content creators, combining audio and video streams to automatically classify events and provide intelligent insights.
Task to create a model that can automatically segment the stomach and intestines on MRI scans.
AI-based reconstruction of subsurface ocean temperature profiles from surface ocean observations using an embedding-based deep learning model.
Application to train a CNN to classify flowers using pytorch and a pretrained model from torchvision and make predictions on new images.
Pretraining and transfer for a 1D Fourier neural operator across PDE families, with controls for budget and learning rate.
An object detection application built with Python and PyTorch using YOLOv5 to identify items, calculate confidence scores, and save labeled visual outputs.
To associate your repository with the pytorc topic, visit your repo's landing page and select "manage topics."