Recommendation Systems
Scaling laws, ranking architectures, representation learning for ranking and retrieval.
Bojian Hou
AI Research Scientist at Meta
Mountain View, California
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.
Recent papers
All 61-
NeurIPS 2026 A Semantic-Sampling Framework for Evaluating Calibration in Open-Ended Question Answering
-
COLM 2026 Beyond Post-Hoc Temperature Scaling: Bilevel Optimization for LLM Calibration
-
KDD 2026 Kunlun: Establishing Scaling Laws for Massive-Scale Recommendation Systems through Unified Architecture Design
-
ICML 2025 Restoring Calibration for Aligned Large Language Models: A Calibration-Aware Fine-Tuning Approach
Background
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. 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.
Published at
Conferences
- NeurIPS
- ICML
- ICLR
- KDD
- AAAI
- IJCAI
- ACM MM
- EMNLP
- COLM
- UAI
- AISTATS
- ICDM
- IEEE BigData
- ACM BCB
- AMIA
- ISBI
- PSB
- MLMI
Journals
- IEEE TNNLS
- IEEE TKDE
- IEEE TAI
- IEEE/ACM TCBB
- Medical Image Analysis
- npj Digital Medicine
- Scientific Reports
- iScience
- Computers in Biology and Medicine
- Innovation in Aging
- Journal of Computer Research and Development
News
All news- Our paper on evaluating calibration in open-ended question answering was accepted at NeurIPS 2026. Paper
- Presented Kunlun at KDD 2026 in Jeju, Korea. Paper
- Our 3D vision-language framework for multi-study CT/MRI report generation was accepted at MLMI 2026. Paper
- Our bilevel optimization approach to LLM calibration was accepted at COLM 2026. Paper
- Kunlun, on scaling laws for massive-scale recommendation systems, was accepted at KDD 2026 (ADS Track). Paper
- IRIS, on interpretable risk clustering for survival analysis, was accepted at IEEE BigData 2025. Paper
Honors and awards
Full list- 2025Best Paper Award, ACM BCB
- 2024Distinguished Paper Award, AMIA Annual Symposium
- 2024PennAITech Innovation Fellow
- 2023Best Paper Award, ACM BCB
- 2021Excellent Doctoral Dissertation Award of Jiangsu Province
- 2021Excellent Doctoral Dissertation Award of Nanjing University
- 2020JSAI Excellent Doctoral Dissertation Award
- 2019CCFAI Outstanding Student Paper Award
- 2017National Scholarship for Ph.D.
Research interests
Browse papers by topicLarge Language Models
Calibration and uncertainty estimation, preference alignment, evaluation, fine-tuning, and LLMs with knowledge graphs.
Trustworthy Machine Learning
Fairness, interpretability, and robustness to evolving features and data streams, from classical models to deep networks.
Optimization and Learning Theory
Bilevel and multi-level optimization, and online learning with regret guarantees.
Multimodal Learning
Vision-language models for 3D imaging, multi-view representation learning with canonical correlation analysis, and deep clustering for survival analysis.
npj Digital Medicine 2026MLMI 2026ACM BCB 2023Medical Image Analysis 2024
Selected publications
All 61 publications* equal contribution† corresponding author
Kunlun: Establishing Scaling Laws for Massive-Scale Recommendation Systems through Unified Architecture Design
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.
A Semantic-Sampling Framework for Evaluating Calibration in Open-Ended Question Answering
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.
Beyond Post-Hoc Temperature Scaling: Bilevel Optimization for LLM Calibration
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.
Restoring Calibration for Aligned Large Language Models: A Calibration-Aware Fine-Tuning Approach
Links post-alignment overconfidence to preference collapse and restores calibration with calibration-aware fine-tuning plus an EM-based ECE regularizer.
Fine-Tuning Attention Modules Only: Enhancing Weight Disentanglement in Task Arithmetic
Fine-tuning only the attention modules improves weight disentanglement for model merging by task arithmetic, without the extra training cost of NTK linearization.
MentalChat16K: A Benchmark Dataset for Conversational Mental Health Assistance
A benchmark that combines synthetic counseling dialogues with anonymized coach and caregiver session transcripts, for building and evaluating conversational mental health assistants.
Fairness-Aware Estimation of Graphical Models
Fair estimation of Gaussian, covariance, and Ising graphical models by jointly minimizing a pairwise graph disparity error and the model loss.
Fair Canonical Correlation Analysis
Learns global CCA projections that close the gap in correlation across protected groups with little loss of accuracy.
Learning with Feature Evolvable Streams
Introduces learning when old features vanish and new ones appear, using an overlap period to keep reusing the old model, with theoretical guarantees.
Experience and education
- 2025–nowMeta
AI Research Scientist, Monetization and Meta Recommendation Systems - 2022–2025University of Pennsylvania
Postdoctoral Fellow, Department of Biostatistics, Epidemiology and Informatics. Advisor: Prof. Li Shen - 2021–2022Cornell University
Postdoctoral Associate - 2020–20214Paradigm
Research Scientist - 2014–2020Nanjing University
Ph.D. in Computer Science, LAMDA Group. Advisor: Prof. Zhi-Hua Zhou - 2010–2014Nanjing University
B.Sc. in Computer Science
Academic service
Full list- OrganizingProgram Chair, VADH Workshop at ICCV 2025
- Senior PCIJCAI 2021 · Top 25% PC member, AAAI 2021
- PC member or reviewerNeurIPS (2018 to 2025), ICML (2019, 2021, 2022, 2024, 2025), ICLR (2019, 2020), KDD (2020, 2026), AAAI (2019 to 2021), AISTATS (2019, 2020)
- Journal reviewerTPAMI, TNNLS, Nature Methods, Medical Image Analysis, Machine Learning, TKDD, Scientific Reports