[ICLR 2026]🚀ReVisual-R1 is a 7B open-source multimodal language model that follows a three-stage curriculum—cold-start pre-training, multimodal reinforcement learning, and text-only reinforcement learning—to achieve faithful, concise, and self-reflective state-of-the-art performance in visual and textual reasoning.

MixGen: A New Multi-Modal Data Augmentation

[ICLR 2025] Official implementation of paper "Improving Data Efficiency via Curating LLM-Driven Rating Systems"

[AAAI 2022] Official Implementation of Active Learning for Domain Adaptation: An Energy-based Approach https://arxiv.org/abs/2112.01406

Official Implementation of DE-DETR and DELA-DETR in "Towards Data-Efficient Detection Transformers"

Data Efficacy for Language Model Training

Official Implementation of DE-CondDETR and DELA-CondDETR in "Towards Data-Efficient Detection Transformers"

This repository contains the pretraining code for the Pillar-0 model.

Code for the "Sample-efficient Integration of New Modalities into Large Language Models" paper

Finetuning framework for Pillar medical imaging models.

This Git repository contains a PowerShell script with a user-friendly interface to automatically organize cluttered files into folders by year and month. Ideal for individuals who struggle with file organization, the tool frees up time and simplifies finding and accessing files.

Official repository for Color Equivariant Convolutional Networks.

Official PyTorch implementation of "GeoPurify: A Data-Efficient Geometric Distillation Framework for Open-Vocabulary 3D Segmentation" (ICLR 2026).

Code base for the paper "Instruction Tuned Models are Quick Learners".

Automatic Label Error Correction www.techrxiv.org/users/679328/articles/731085

🌿 dPrune: A Framework for Data Pruning

[CVPR '26] Learnability-Driven Submodular Optimization for Active Roadside 3D Detection

An interactive tool for generating compact, representative coresets from images, enabling faster, memory-efficient processing for tasks like segmentation and clustering with minimal accuracy loss.

Efficient image segmentation using KMeans clustering with coreset sampling to speed up processing while preserving segmentation quality.

Confluence Labs is a San Francisco AI research company (Y Combinator Winter 2026, backed by Paul Graham) focused on learning efficiency — building machine learning systems that learn from limited data and few experiments.