Publications - Yao Lai

ICML

LithoGRPO: Fast Inverse Lithography via GRPO Reinforced Flow Matching

Yao Lai, Xuyuan Xiong, Zeyue Xue, Guojin Chen, Jing Wang, Xihui Liu, Rui Zhang, Robert Mullins, Bei Yu, Ping Luo

International Conference on Machine Learning (ICML) 2026

LithoGRPO is an inverse lithography framework that unifies flow matching with GRPO-based reinforcement learning to optimize masks under physics-based rewards, using a fast shot-counting algorithm (over 130x speedup) to reach state-of-the-art performance.

[Paper] [Code]

ICML

LithoGRPO: Fast Inverse Lithography via GRPO Reinforced Flow Matching

Yao Lai, Xuyuan Xiong, Zeyue Xue, Guojin Chen, Jing Wang, Xihui Liu, Rui Zhang, Robert Mullins, Bei Yu, Ping Luo

International Conference on Machine Learning (ICML) 2026

LithoGRPO is an inverse lithography framework that unifies flow matching with GRPO-based reinforcement learning to optimize masks under physics-based rewards, using a fast shot-counting algorithm (over 130x speedup) to reach state-of-the-art performance.

[Paper] [Code]

TCAD

AnalogCoder-Pro: Unifying Analog Circuit Generation and Optimization via Multi-modal LLMs

Yao Lai, Sungyoung Lee, Guojin Chen, Souradip Poddar, Mengkang Hu, Bei Yu, Ping Luo, David Z. Pan

IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems 2026

AnalogCoder-Pro is a multimodal LLM framework that automates analog circuit design by unifying topology generation and device sizing through a diagnosis-and-repair feedback loop, reusable circuit library, and Bayesian optimization, outperforming existing methods on a 13-circuit-type benchmark.

[Paper] [Code]

TCAD

AnalogCoder-Pro: Unifying Analog Circuit Generation and Optimization via Multi-modal LLMs

Yao Lai, Sungyoung Lee, Guojin Chen, Souradip Poddar, Mengkang Hu, Bei Yu, Ping Luo, David Z. Pan

IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems 2026

AnalogCoder-Pro is a multimodal LLM framework that automates analog circuit design by unifying topology generation and device sizing through a diagnosis-and-repair feedback loop, reusable circuit library, and Bayesian optimization, outperforming existing methods on a 13-circuit-type benchmark.

[Paper] [Code]

NeurIPS

FUDOKI: Discrete Flow-based Unified Understanding and Generation via Kinetic-Optimal Velocities

Jin Wang*, Yao Lai*, Aoxue Li, Shifeng Zhang, Jiacheng Sun, Chengyue Wu, Zhenguo Li, Ping Luo (* equal contribution)

Conference on Neural Information Processing Systems (NeurIPS) 2025 Spotlight

FUDOKI is a novel unified multimodal model that replaces traditional autoregressive architectures with discrete flow matching, enabling more flexible and effective visual understanding and image generation with performance comparable to state-of-the-art models.

[Paper] [Project] [Code]

NeurIPS

FUDOKI: Discrete Flow-based Unified Understanding and Generation via Kinetic-Optimal Velocities

Jin Wang*, Yao Lai*, Aoxue Li, Shifeng Zhang, Jiacheng Sun, Chengyue Wu, Zhenguo Li, Ping Luo (* equal contribution)

Conference on Neural Information Processing Systems (NeurIPS) 2025 Spotlight

FUDOKI is a novel unified multimodal model that replaces traditional autoregressive architectures with discrete flow matching, enabling more flexible and effective visual understanding and image generation with performance comparable to state-of-the-art models.

[Paper] [Project] [Code]

AAAI

AnalogCoder: Analog Circuit Design via Training-Free Code Generation

Yao Lai, Sungyoung Lee, Guojin Chen, Souradip Poddar, Mengkang Hu, David Z. Pan, Ping Luo

AAAI Conference on Artificial Intelligence (AAAI) 2025 OralAAAI 2025 Top-15 Influential Papers (Rank 11)

AnalogCoder is a training-free LLM agent for analog circuit design, using feedback-driven prompts and a circuit library to achieve high success rates, outperforming GPT-4o by designing various circuits.

[Paper] [Code]

AAAI

AnalogCoder: Analog Circuit Design via Training-Free Code Generation

Yao Lai, Sungyoung Lee, Guojin Chen, Souradip Poddar, Mengkang Hu, David Z. Pan, Ping Luo

AAAI Conference on Artificial Intelligence (AAAI) 2025 OralAAAI 2025 Top-15 Influential Papers (Rank 11)

AnalogCoder is a training-free LLM agent for analog circuit design, using feedback-driven prompts and a circuit library to achieve high success rates, outperforming GPT-4o by designing various circuits.

[Paper] [Code]

NeurIPS

Scalable and Effective Arithmetic Tree Generation for Adder and Multiplier Designs

Yao Lai, Jinxin Liu, David Z. Pan, Ping Luo

Conference on Neural Information Processing Systems (NeurIPS) 2024 Spotlight

This work uses reinforcement learning to optimize adder and multiplier designs as tree generation tasks, achieving up to 49% faster speed and 45% smaller size, with scalability to 7nm technology.

[Paper] [Code] [Video]

NeurIPS

Scalable and Effective Arithmetic Tree Generation for Adder and Multiplier Designs

Yao Lai, Jinxin Liu, David Z. Pan, Ping Luo

Conference on Neural Information Processing Systems (NeurIPS) 2024 Spotlight

This work uses reinforcement learning to optimize adder and multiplier designs as tree generation tasks, achieving up to 49% faster speed and 45% smaller size, with scalability to 7nm technology.

[Paper] [Code] [Video]

ICML

ChiPFormer: Transferable Chip Placement via Offline Decision Transformer

Yao Lai, Jinxin Liu, Zhentao Tang, Bin Wang, Jianye Hao, Ping Luo

International Conference on Machine Learning (ICML) 2023

ChiPFormer is an offline RL-based method that achieves 10x faster chip placement with superior quality and transferability to unseen circuits.

[Paper] [Project] [Code]

ICML

ChiPFormer: Transferable Chip Placement via Offline Decision Transformer

Yao Lai, Jinxin Liu, Zhentao Tang, Bin Wang, Jianye Hao, Ping Luo

International Conference on Machine Learning (ICML) 2023

ChiPFormer is an offline RL-based method that achieves 10x faster chip placement with superior quality and transferability to unseen circuits.

[Paper] [Project] [Code]

NeurIPS

MaskPlace: Fast Chip Placement via Reinforced Visual Representation Learning

Yao Lai, Yao Mu, Ping Luo

Conference on Neural Information Processing Systems (NeurIPS) 2022 Spotlight

MaskPlace is a method that leverages pixel-level visual representation for chip placement, achieving superior performance with simpler rewards, 60%-90% wirelength reduction, and zero overlaps.

[Paper] [Code] [Project] [Video]

NeurIPS

MaskPlace: Fast Chip Placement via Reinforced Visual Representation Learning

Yao Lai, Yao Mu, Ping Luo

Conference on Neural Information Processing Systems (NeurIPS) 2022 Spotlight

MaskPlace is a method that leverages pixel-level visual representation for chip placement, achieving superior performance with simpler rewards, 60%-90% wirelength reduction, and zero overlaps.

[Paper] [Code] [Project] [Video]

ECAI

OpenSMax: Unknown Domain Generation Algorithm Detection

Yao Lai, Guolou Ping, Yuexin Wu, Chenhui Lu, Xiaojun Ye

European Conference on Artificial Intelligence (ECAI) 2020

We propose OpenSMax, a LSTM-based model that learns patterns from domain names to detect botnet-generated domains and flag previously unseen DGA families as “unknown,” improving accuracy over prior methods while keeping false alarms under control.

[Paper] [Code]

ECAI

OpenSMax: Unknown Domain Generation Algorithm Detection

Yao Lai, Guolou Ping, Yuexin Wu, Chenhui Lu, Xiaojun Ye

European Conference on Artificial Intelligence (ECAI) 2020

We propose OpenSMax, a LSTM-based model that learns patterns from domain names to detect botnet-generated domains and flag previously unseen DGA families as “unknown,” improving accuracy over prior methods while keeping false alarms under control.

[Paper] [Code]