Hello!

Recent Updates

• [2026-09-01] I have joined Singapore-MIT Alliance of Research and Technology (SMART), under M3S IRG, as a Postdoctorate Associate.


• [2026-06-01] I have several accepted workshop papers at ICML 2026 DEMO workshop. They are all about how Bayesian Optimization can be used for LLM training. Check them out!


• [2025-12-20] I am on the Honour List of Student Tutors for Excellence in Teaching at the School of Computing, NUS.


• [2025-07-08] I present "PIED: Physics-Informed Experimental Design for Inverse Problems" at the AI4X Conference 2025 in Singapore.


• [2025-01-23] My co-first author paper titled "PIED: Physics-Informed Experimental Design for Inverse Problems" has been accepted to ICLR 2025.

For archive of all past updates, see here.

For more details about my current progress see my academic works and my CV.

About Myself

Photo of myself (10 points if you know where this is)

Hello. I am Apivich Hemachandra (in Thai: อภิวิชญ์ ​เหมะจันทร), although I also go by my nickname¹ “Kaotoo” (in Thai: ข้าวตู).

I am currently a Postdoctorate Associate at Singapore-MIT Alliance for Research and Technology (SMART), under the M3S IRG. Previously, I was a PhD student at the School of Computing at National University of Singapore (NUS), under the supervision of See-Kiong Ng and Bryan Low Kian Hsiang.

My general research interest is on data-efficient ML, especially on active learning (AL) and black-box optimisation (BBO). In particular, I am interested in two directions:

1. How AL and BBO can be used to improve the efficiency of NN training, particularly through the lens of training data, NN architecture, prior knowledge, and compute. These training components may also have inter-dependencies, making the optimal choice for them more difficult to determine. Because it’s expensive to try out many potential NN training setup (especially LLMs), we need to be a bit clever with what training setup we may want to try running.

2. How we can evolve AL and BBO to be more efficient through the use of existing prior knowledge and/or LLMs. This is particularly relevant in scientific domains, where we may have additional information which can be used to reduce how many experiments/simulations we need for AL or BBO. Because of the different form the prior knowledge can take up, and with the different choices we can make during experimentation/simulation, we may want to use cleverer systems to make these decisions for us rather than doing it manually.

Prior to my PhD, I completed my Bachelor’s degree from Mahidol University International College (MUIC) in Thailand. I majored in Physics, however I also completed minors in Computer Science and Mathematics. My Bachelor’s Thesis was related to data diversification and submodular maximisation². I have also worked on different computer science-related projects with researchers in Thailand and in collaboration with local firms as a data analyst.

Personal Life

A view of Chiang Mai from a rooftop (click image to enlarge)

I was born in Bangkok, however spent pretty much all of my childhood in Chiang Mai, right up until the end of high school. I then spent four years in Bangkok/Nakhon Pathom to complete my undergraduate degree. I have only been in Singapore since the start of my PhD (in 2021).

When I am not busy doing work, I enjoy playing and listening to music. I am a mediocre drummer, bassist and vocalist, and have been playing varying amounts of music since high school. My music taste isn’t that varied (generally alternative, indie or some genres of pop), but I do discuss them somewhere on this website (you’ll have to find it yourself though). If you meet me in person, feel free to discuss music or recent live gigs you’ve been to with me.

I enjoy watching football (or soccer as some may call it), and am a fan of Nottingham Forest (who will hopefully be are now in the Premier League soon).

I also make maths videos whenever I have enough free time.