Institution profile

Singapore-MIT Alliance for Research and Technology

Academic institutionasia · sg
Official website
Research library64linked papers
Opportunities0open roles
Selected work

Representative Papers

MineDraft: A Framework for Batch Parallel Speculative Decoding

Feb 24, 2026

Standard speculative decoding suffers from limited inference acceleration due to the strict serial dependency between draft generation and verification. This work proposes MineDraft, a novel batch-parallel speculative decoding framework that introduces a dual-batch pipelined scheduling mechanism, overlapping draft generation for one batch of requests with the verification phase of another. Furthermore, MineDraft incorporates a cooperative verification protocol between the draft and target models to ensure output correctness while significantly improving hardware utilization. Integrated into the vLLM system, the proposed approach reduces end-to-end latency by up to 39% and achieves a throughput improvement of up to 75% compared to standard speculative decoding.

2 citationsRead paper

Understanding Interfirm AI Talent Flow Networks through Online Professional Profiles

Oct 06, 2026

This study addresses the limitation of existing research that predominantly conceptualizes AI capabilities as firm-internal resources, overlooking their dynamic inter-organizational distribution through talent mobility. Leveraging 535 million employment records, we construct a global AI talent mobility network and integrate complex network analysis with event study methodology to examine how network centrality influences firm value. Our findings reveal that while AI talent inflows are highly concentrated, core network positions remain contestable. Furthermore, network centrality exhibits a significant positive association with firm value, and upward positional shifts presage subsequent value enhancement. This research demonstrates that network structures encode economic information beyond mere labor scale, offering a novel perspective for understanding the cross-organizational spillovers of AI capabilities.

0 citationsRead paper

Evolving in Thought Space: Training a Small Model at Test Time Unlocks Better Discoveries

Oct 05, 2026

This study addresses the high computational costs of test-time training for large language models and the credit assignment challenges arising from coupled policy and implementation. We propose Guidance-TTT, a framework that introduces a novel guidance-execution decoupled architecture. Specifically, it freezes a large model to handle code implementation while exclusively training a small model to optimize high-level decision-making policies, thereby restricting test-time learning to short-horizon decision sequences. Furthermore, we design an adaptive group-relative reinforcement learning objective that leverages verifier feedback to enable online policy updates. Experimental results demonstrate that, in offline settings, this framework outperforms state-of-the-art methods across four major domains, including combinatorial optimization and machine learning, while significantly reducing computational overhead and enhancing exploration efficiency.

0 citationsRead paper

Plan-to-Synthesis: Cross-City Human Mobility Generation via Semantic Latent Flow Matching

Sep 26, 2026

This study addresses the poor generalizability and high transfer costs of single-city models in cross-city trajectory generation by proposing the SeMoFlow framework. This method pioneers a hierarchical semantic ID encoding scheme for heterogeneous points of interest (POIs) to construct a shared representation space, and adopts a "planning-synthesis" hierarchical paradigm. Specifically, an autoregressive planner generates macroscopic trajectory structures, while latent flow matching combined with decoding grounding techniques achieves high-fidelity microscopic synthesis. Experiments demonstrate that SeMoFlow significantly outperforms existing baselines on large-scale multi-city datasets. By effectively preserving city-specific patterns, the proposed framework enables efficient joint generation and seamless cross-city transfer.

0 citationsRead paper
Recent publications

Latest Papers

Understanding Interfirm AI Talent Flow Networks through Online Professional Profiles

Oct 06, 2026

This study addresses the limitation of existing research that predominantly conceptualizes AI capabilities as firm-internal resources, overlooking their dynamic inter-organizational distribution through talent mobility. Leveraging 535 million employment records, we construct a global AI talent mobility network and integrate complex network analysis with event study methodology to examine how network centrality influences firm value. Our findings reveal that while AI talent inflows are highly concentrated, core network positions remain contestable. Furthermore, network centrality exhibits a significant positive association with firm value, and upward positional shifts presage subsequent value enhancement. This research demonstrates that network structures encode economic information beyond mere labor scale, offering a novel perspective for understanding the cross-organizational spillovers of AI capabilities.

0 citationsRead paper

Evolving in Thought Space: Training a Small Model at Test Time Unlocks Better Discoveries

Oct 05, 2026

This study addresses the high computational costs of test-time training for large language models and the credit assignment challenges arising from coupled policy and implementation. We propose Guidance-TTT, a framework that introduces a novel guidance-execution decoupled architecture. Specifically, it freezes a large model to handle code implementation while exclusively training a small model to optimize high-level decision-making policies, thereby restricting test-time learning to short-horizon decision sequences. Furthermore, we design an adaptive group-relative reinforcement learning objective that leverages verifier feedback to enable online policy updates. Experimental results demonstrate that, in offline settings, this framework outperforms state-of-the-art methods across four major domains, including combinatorial optimization and machine learning, while significantly reducing computational overhead and enhancing exploration efficiency.

0 citationsRead paper

Plan-to-Synthesis: Cross-City Human Mobility Generation via Semantic Latent Flow Matching

Sep 26, 2026

This study addresses the poor generalizability and high transfer costs of single-city models in cross-city trajectory generation by proposing the SeMoFlow framework. This method pioneers a hierarchical semantic ID encoding scheme for heterogeneous points of interest (POIs) to construct a shared representation space, and adopts a "planning-synthesis" hierarchical paradigm. Specifically, an autoregressive planner generates macroscopic trajectory structures, while latent flow matching combined with decoding grounding techniques achieves high-fidelity microscopic synthesis. Experiments demonstrate that SeMoFlow significantly outperforms existing baselines on large-scale multi-city datasets. By effectively preserving city-specific patterns, the proposed framework enables efficient joint generation and seamless cross-city transfer.

0 citationsRead paper