also published as Bo-Jian Hou · 侯博建
I am an AI Research Scientist at Meta in Monetization and Meta Recommendation Systems (MRS). My work centers on LLM-style model architectures and scaling laws for massive-scale ads ranking (e.g., Kunlun, KDD 2026) and on representation learning that connects organic and ads signals for ranking and retrieval. I also continue my research on trustworthy large language models, especially calibration, uncertainty estimation, and preference alignment.
Before joining Meta, I was a postdoctoral fellow in the Department of Biostatistics, Epidemiology and Informatics at the University of Pennsylvania, advised by Prof. Li Shen, where I led a Trustworthy AI research line and mentored 20+ students. Earlier, I was a postdoctoral associate at Cornell University and a research scientist at 4Paradigm. I received my B.Sc. and Ph.D. in Computer Science from Nanjing University in 2014 and 2020, supervised by Prof. Zhi-Hua Zhou in the LAMDA Group.
I am interested in both the theoretical foundations and the real-world deployment of machine learning, and in building AI systems that are technically rigorous, trustworthy, and socially responsible.
NeurIPS (5) · ICML · ICLR · KDD (2) · COLM · AAAI · IJCAI · UAI · AISTATS · EMNLP Findings · IEEE TNNLS (3) · IEEE TKDE (2) · Medical Image Analysis · npj Digital Medicine
Best Paper Award, ACM BCB 2025 and 2023 · Distinguished Paper Award, AMIA 2024 · Oral, KDD 2025 · Excellent Doctoral Dissertation Award, Jiangsu Province
My research spans algorithms, theory, and applications of machine learning. Representative papers are linked under each area; the publications page can be filtered by topic.
Scaling laws, ranking architectures, representation learning for ranking and retrieval.
Calibration and uncertainty estimation, preference alignment, evaluation, fine-tuning.
Fairness learning and interpretability, from classical models to deep networks.
Feature evolvable learning, online learning, bilevel optimization, semi-supervised learning.
Alzheimer's disease and dementia, mental health, medical imaging and vision-language models.
KDD 2025 · Medical Image Analysis 2024 · npj Digital Medicine 2026
* equal contribution · † corresponding author
A recommendation architecture designed for predictable power-law scaling: it raises model FLOPs utilization from 17% to 37% on NVIDIA B200 GPUs, doubles scaling efficiency, and is deployed in major Meta Ads models.
Sem-ECE evaluates LLM calibration in open-ended QA by sampling answers, grouping them into semantic classes, and using class frequencies as confidence; two estimators are analyzed and tested on five commercial LLMs.
Replaces post-hoc temperature scaling with a bilevel objective that maximizes predictive entropy, solved by a first-order method that scales to LLMs; calibration gains are strongest out of domain.
Links post-alignment overconfidence to preference collapse and restores calibration with calibration-aware fine-tuning plus an EM-based ECE regularizer.
Fine-tuning only the attention modules improves weight disentanglement for model merging by task arithmetic, without the extra training cost of NTK linearization.
A benchmark that combines synthetic counseling dialogues with anonymized coach and caregiver session transcripts, for building and evaluating conversational mental health assistants.
Fair estimation of Gaussian, covariance, and Ising graphical models by jointly minimizing a pairwise graph disparity error and the model loss.
Learns global CCA projections that close the gap in correlation across protected groups with little loss of accuracy.
Introduces learning when old features vanish and new ones appear, using an overlap period to keep reusing the old model, with theoretical guarantees.