LC-JEPA: What Pretraining Objectives Keep and Discard

github Changing-look AGN (CLAGN) are quasars that switch spectral type over months to years — direct evidence of an accretion-state transition around a supermassive black hole. They’re rare (~1,000 known) and almost all of them were found by accident, in spectral repeat visits nobody scheduled specifically to catch one. If the transition leaves a signature in the photometry — the light curve alone, no spectrum required — you could search for CLAGN systematically across every quasar a survey has ever imaged, instead of waiting to get lucky....

September 2026 · Yash Thube

paper implementations

PyTorch implementations of research papers, aimed at deepening my understanding of the underlying concepts.

Yash Thube (2025)

SSL-Vision

Implementation of four self-supervised learning (SSL) algorithms; SimCLR, MoCo-v2, BYOL, and DINO on the CIFAR-10 dataset using a ResNet-18 backbone. All scripts are designed to run on Google Colab Free Tier (single GPU, ~12 GB RAM) with CIFAR-10.

Yash Thube

Task-Aware Segment Anything with LoRA

Abstract PyTorch pipeline that uses a hypernetwork to generate task-specific LoRA adapters for Meta’s Segment Anything Model from natural language prompts, targeted segmentation on COCO instances and benchmarked mIoU via pycocotools. Links You can find the repo here.

Yash Thube