Chia-Hsiang Kao

Image

Email / CV / Google Scholar

I am a 3rd-year CS PhD student at Cornell, studying multi-modal reasoning and learning, advised by Prof. Bharath Hariharan. My goal is to deepen our understanding of multimodal models and build reliable, interpretable ML systems.

This past summer, I was an Applied Scientist Intern at Amazon, and I am now continuing my industry experience as a Student Researcher at Meta for the 2025-2026 academic year.

Before joining Cornell, I obtained my Medical Doctor degree in Taiwan.


Publications
Publication preview
Towards LLM Agents for Earth Observation
Image Image Image
Chia-Hsiang Kao, Wenting Zhao, Shreelekha Revankar, Samuel Speas, Snehal Bhagat, Rajeev Datta, Cheng Perng Phoo, Utkarsh Mall, Carl Vondrick, Kavita Bala, Bharath Hariharan
We ask and answer: Are AI systems ready for reliable Earth Observation?
Publication preview
Counter-Current Learning: A Biologically Plausible Dual Network Approach for Deep Learning
Image Image Image
Chia-Hsiang Kao, Bharath Hariharan
We design a non-backpropagation learning algorithm that mimicks the counter-current exchange mechanisms observed in biological systems.
Publication preview
AllClear: A Comprehensive Dataset and Benchmark for Cloud Removal in Satellite Imagery
Image Image Image Image
Hangyu Zhou*, Chia-Hsiang Kao*, Cheng Perng Phoo, Utkarsh Mall, Bharath Hariharan, Kavita Bala
We build up the largest collection of satellite images with cloud occlusions.
Publication preview
Caduceus: Bi-directional equivariant long-range dna sequence modeling
Image Image Image Image
Yair Schiff, Chia-Hsiang Kao, Aaron Gokaslan, Tri Dao, Albert Gu, Volodymyr Kuleshov
We design a DNA foundation model, specialized in long-context sequence modeling, combining the bi-directional equivariant intuition.
Publication preview
FedBug: A Bottom-Up Gradual Unfreezing Framework for Federated Learning
Image Image Image
Chia-Hsiang Kao, Yu-Chiang Frank Wang
We design a federated learning framework to mitigate the dataset shift by gradually unfreezing the model.
Publication preview
MAML Is a Noisy Contrastive Learner in Classification
Image Image Image Image
Chia-Hsiang Kao, Wei-Chen Chiu, Pin-Yu Chen
We show that Model-agnostic meta-learning (MAML) acts as contrastive leaerning.
Awards and Scholarships
2025
Foundational AI PhD Fellow, Cornell University
2021
Student Travel Award, MICCAI
2020
Undergraduate Research Fellowship, National Science and Technology Council, Taiwan
2018
Undergraduate Research Fellowship, National Science and Technology Council, Taiwan
2018
Summer Research Fellowship, National Science and Technology Council, Taiwan
Services
Conference Reviewer
AAAI[25], AISTATS[25], AutoML[22], ICML[25], ICLR[25], NeurIPS[21,24,25], ICLR [25,26]
Interests
🏊🏼 and 🏄
Got a lifeguard license at 18!
🏃 and 🏞️
One half-marathon, two 10Ks, and five 8.9Ks so far!
🎤 and 🎸
Used to play the viola, but play the guitar more now!

Template: this
Last update: Nov. 2024