Tian Lan

About Me

I am currently an Algorithm Expert at Alibaba Token Hub (ATH) – MaaS, where I focus on automatic evaluation and long-horizon agent technologies, driving both frontier research and real-world business applications. I joined Alibaba through the 2025 Alistar (阿里星) campus recruitment program. I received my Ph.D. in Computer Science and Technology from Beijing Institute of Technology (BIT), advised by Prof. Xian-Ling Mao (毛先领). During my internships, I was primarily mentored by Yan Wang (王琰).

My work spans both academia and industry: I have published at top-tier AI venues including ICLR, NeurIPS, ICML, ACL, EMNLP, ACM MM, SIGKDD, and TOIS, with representative works such as Copy is All You Need, Block-Attention, and Contrastive Search, accumulating over 1,700 citations on Google Scholar. I regularly serve as a reviewer for top-tier conferences and journals, including ICLR, AAAI, NeurIPS, ACL ARR, ICML, and TOIS.

🚀 We're hiring! Our team is actively recruiting both daily interns and campus-recruitment interns. If you're a self-motivated student excited about exploring how agents can be applied to real-world problems, feel free to reach out — I'd love to hear from you.

I'm always happy to discuss research on long-horizon agents, automatic evaluation, and large language models. Feel free to reach out via email if you'd like to collaborate.

Recent Highlights

  • Jul. 2026. Two papers accepted to ACM MM 2026, covering multi-modal understanding and generation.
  • Jul. 10th, 2026. HSCodeComp, our challenging benchmark on agents' ability to apply complex expert-domain rules, won the ACL 2026 Best Resource Award.
  • Apr. 30th, 2026. One paper accepted to ICML 2026, on agentic memory.
  • Apr. 4th, 2026. Five papers accepted to ACL 2026, covering agent systems, reasoning, and more.
  • Mar. 2026. Released Marco DeepResearch Agent, the best agentic search model at the 8B scale, built on the Marco DeepResearch project.
  • Nov. 2025. One paper accepted to AAAI 2026, on long-document generation.
  • Oct. 24th, 2025. Launched the Marco DeepResearch project, an open agentic-research framework covering agentic harness, memory, and search.
  • Oct. 23rd, 2025. Released DeepWideSearch, a benchmark for long-horizon agentic search.
  • Aug. 2025. One paper accepted to EMNLP 2025, on enhancing LLMs' self-critique ability.
  • Jun. 30th, 2025. One paper accepted to NLPCC 2025, on automatic peer review.
  • Jun. 20th, 2025. Successfully defended my dissertation and earned my Ph.D. in Computer Science and Technology from Beijing Institute of Technology.
  • Mar. 27th, 2025. One paper, Decider, accepted to IEEE TKDE, on rule-controllable decoding for language generation.
  • Sep. 2024. One paper, CriticEval, accepted to NeurIPS 2024.
  • May. 25th, 2024. One survey paper on automatic evaluation of generated text accepted to CCL 2024, forming part of the literature foundation for my Ph.D. dissertation.
  • May. 16th, 2024. One paper accepted to ACM SIGKDD 2024, on retrieval-augmented LLMs for scientific papers.
  • Nov. 6th, 2023. One paper accepted to ACM TOIS, on dense retrieval for dialogue response selection.
  • Sep. 22nd, 2023. One paper accepted to NeurIPS 2023, on understanding neural text degeneration.
  • Jul. 21st, 2023. One paper, PandaGPT, accepted to the ACL-affiliated TLLM 2023 workshop; now cited 450+ times.
  • May 8th, 2023. One paper accepted to ACM TOIS, on efficient coarse-grained dialogue response selection.
  • Jan. 2023. One paper accepted to ICLR 2023, on retrieval-augmented text generation.
  • Sep. 14th, 2022. One paper accepted to NeurIPS 2022 (Spotlight), on contrastive text generation.
  • Feb. 23rd, 2022. One paper accepted to ACL 2022, on cross-lingual phrase retrieval.
  • Sep. 3rd, 2021. One paper, LASH, accepted to IEEE TKDE, on large-scale academic semantic hashing.
  • Nov. 13th, 2020. One paper, PONE, accepted to ACM TOIS, on automatic evaluation for open-domain dialogue systems.

Education

  • Ph.D. in Computer Science and Technology (Combined Master–Ph.D. Program) Beijing Institute of Technology Advisor: Prof. Xian-Ling Mao (毛先领)
    2019 – 2025
  • B.Eng. in Computer Science Beijing Institute of Technology
    2015 – 2019

Experience

Selected Publications Google Scholar ↗

* indicates equal contribution.

Projects

Marco DeepResearch

Alibaba's flagship agentic research effort on DeepResearch agents, covering agentic harness, agentic memory, self-evolving skills, and agentic search.

AgentDeepResearchAgentic Search
GitHub

Contrastive Search / SimCTG

A decoding algorithm for neural text generation, later integrated into HuggingFace transformers (v4.24.0) as a built-in decoding method and featured in an official HF blog post.

DecodingHuggingFace
GitHub 🤗 HF Blog

Effidit

Tencent AI Lab's AI-powered writing assistant product; contributed during my research internship on text generation and retrieval-augmented generation.

ContributorProduct
Website

HammerLLM

A 1.4B-parameter efficient LLM pre-training framework built by the DataHammer Lab, covering data pipeline, training, and evaluation for academic-scale LLM research.

Pre-trainingLLM
GitHub

Mozi-LLM (science-llm)

Retrieval-augmented LLM toolkit for scientific literature understanding, the open-source companion to our SIGKDD 2024 paper on hierarchical context augmentation.

RAGScientific NLP
GitHub

InternLM2

An open-sourced large language model that outperforms its predecessors in comprehensive evaluations across multiple dimensions, notably in long-context modeling and reasoning.

ContributorLLM
arXiv

PandaGPT

One model to instruction-follow them all: a general-purpose instruction-following model that combines the multimodal encoders of ImageBind with the reasoning power of large language models. Accepted to the ACL-affiliated TLLM 2023 workshop; 450+ citations on Google Scholar.

ContributorMultimodal
Website

Awards & Honors

  • ACL 2026 Best Resource Award — for HSCodeComp.
  • Alistar (阿里星) — Alibaba Group. 2025.
  • National Scholarship (国家奖学金) — Beijing Institute of Technology. 2024.
  • Tencent Qingyun Program (腾讯青云计划) — Tencent. 2025.
  • Tencent Rhino-Bird Elite Intern Program (腾讯犀牛鸟计划菁英实习生) — Tencent. 2022.
  • National Scholarship (国家奖学金) — Beijing Institute of Technology. 2017.

Talks

  • ATH-MaaS Team ACL 2026 Best Resource Paper Sharing: A New Benchmark for Expert-level Deep Search Agents in Real-world Scenarios — Alibaba Internal Algorithm Evening Seminar (算法晚自习). 2026/08.
  • DeepResearch Agent — Invited talk (Alibaba Cloud) at ACL 2026. 2026/07/08. [slides]
  • Agent Harness Practice Guide: Vertical Harness Design for Deep & Broad Search Tasks — Alibaba Internal Algorithm Evening Seminar (算法晚自习). 2026/04.
  • UMEM — Invited talk at the China Generative AI Summit 2026. 2026/04. [slides]
  • Agentic Memory — Invited talk hosted by Dongfang Li, Harbin Institute of Technology, Shenzhen. 2026/04.
  • DeepResearch Agent at the 1st CCF Conference on Large Language Models & AI Engineering. 2025/04. [slides]
  • Copy is All You Need — Invited talk hosted by Yutao Zhu, Gaoling School of Artificial Intelligence, Renmin University of China. 2024/03.
  • Repetition in Repetition Out (NeurIPS 2023) — Invited talk hosted by Ruihua Song, Gaoling School of Artificial Intelligence, Renmin University of China. 2023/10.

Contact

The best way to reach me is by email at lantiangmftby@gmail.com. I'm also on GitHub and Google Scholar.

Visitors

A live map of visitors from around the world.

Site visits: - · Unique visitors: -

© Tian Lan. Built with plain HTML/CSS/JS, hosted on GitHub Pages.