Unifying Video Tasks via Spatiotemporal Analogy

Chia-Hsiang Kao, Belinda Zeng, Bharath Hariharan, Menglin Jia

Under review[arXiv][Project]

A text-free framework that executes and generalizes to diverse, unseen video tasks by completing a single visual analogy on a spatiotemporal canvas.

Δynamics: Language-Based Representation for Inferring Rigid-Body Dynamics From Videos

Chia-Hsiang Kao, Cong Phuoc Huynh, Chien-Yi Wang, Noranart Vesdapunt, Stefan Stojanov, Bharath Hariharan, Oleksandr Obiednikov, Ning Zhou

CVPR 2026[arXiv][Project]

A vision-language model that uses language as a unified representation to infer rigid-body dynamics from video, bridging perception and physics simulation.

Towards LLM Agents for Earth Observation

Chia-Hsiang Kao, Wenting Zhao, Cheryl Lam, Aarush Umap, Shreelekha Revankar, Samuel Speas, Snehal Bhagat, Rajeev Datta, Cheng Perng Phoo, Utkarsh Mall, Carl Vondrick, Kavita Bala, Bharath Hariharan

ACL 2026 Findings[arXiv][Project]

Are AI agents ready for reliable Earth observation? A benchmark for code-executing agents on real satellite data.

Counter-Current Learning: A Biologically Plausible Dual Network Approach for Deep Learning

Chia-Hsiang Kao, Bharath Hariharan

NeurIPS 2024[arXiv][Code]

A non-backpropagation learning algorithm inspired by counter-current exchange in biological systems.

AllClear: A Comprehensive Dataset and Benchmark for Cloud Removal in Satellite Imagery

Hangyu Zhou*, Chia-Hsiang Kao*, Cheng Perng Phoo, Utkarsh Mall, Bharath Hariharan, Kavita Bala

NeurIPS 2024 Datasets & Benchmarks[arXiv][Project][Code]

The largest collection of satellite images with cloud occlusions, with a benchmark for cloud removal.

MAML Is a Noisy Contrastive Learner in Classification

Chia-Hsiang Kao, Wei-Chen Chiu, Pin-Yu Chen

ICLR 2022[arXiv][Code][Blog]

We show that model-agnostic meta-learning (MAML) acts as contrastive learning.