Sohee Yang (양소희)

Sohee Yang (양소희)

PhD Student/Research Scientist Intern

UCL NLP

Google DeepMind

I am a second-year PhD student at UCL and a part-time research scientist intern at Google DeepMind, splitting my time between the two organizations during the course of my PhD. I am co-advised by Prof. Pontus Stenetorp and Prof. Sebastian Riedel at UCL NLP Group. I obtained my Master’s degree at Kim Jaechul Graduate School of AI, KAIST, under the guidance of Prof. Minjoon Seo at Language & Knowledge Lab. Prior to this, I was a research engineer at Naver Clova. My research interests lie in the ways to understand and enhance the reasoning ability of NLP/ML systems. Curiosity drives me, and I am happy to follow wherever it leads me.

I have suffered from severe RSI (Repetitive Strain Injury) on all my fingers developed from keyboard overuse from Jan 2021 to July 2022; during this period, I could not type anything without seriously aggravating nerve pain. Thankfully, in July 2022, the RSI finally began to be healed through numerous sessions of TPIs (Trigger Point Injections) on my arms by a very skillful doctor and I could restart leading a research project in August 2022. I plan to post on my homepage someday about what have been helpful for my treatment, in case it is helpful for others suffering from RSI as well.

Last page update: Sep 26, 2024

Interests

  • Large Language Models
  • Natural Language Processing
  • Machine Learning

Education

  • PhD in Computer Science, March 2023 - Present

    University College London (UCL)

  • MS in Artificial Intelligence, March 2021 - Feb 2023

    Kim Jaechul Graduate School of AI, KAIST

  • BS in Computer Science and Engineering (Summa Cum Laude), March 2014 - Feb 2018

    Handong Global University (GPA: 4.45/4.5, 1st rank in CSEE)

Industry Experience

Publications

(2024). How Do Large Language Models Acquire Factual Knowledge During Pretraining?. In NeurIPS 2024.

PDF

(2024). Hopping Too Late: Exploring the Limitations of Large Language Models on Multi-Hop Queries. In EMNLP 2024.

PDF Code

(2024). Exploring the Practicality of Generative Retrieval on Dynamic Corpora. In EMNLP 2024.

PDF

(2024). Do Large Language Models Latently Perform Multi-Hop Reasoning?. In ACL 2024.

PDF Poster Slides

(2023). Improving Probability-based Prompt Selection Through Unified Evaluation and Analysis. TACL 2024 (presented in ACL 2024).

PDF Code Dataset Poster Slides

(2023). Investigating the Effectiveness of Task-Agnostic Prefix Prompt for Instruction Following. In AAAI 2024.

PDF Code

(2022). Knowledge Unlearning for Mitigating Privacy Risks in Language Models. In ACL 2023.

PDF Code

(2022). Contextualized Generative Retrieval. In Findings of ACL 2023.

PDF Code

(2022). Generative Multi-hop Retrieval. In EMNLP 2022.

PDF Code

(2022). TemporalWiki: A Lifelong Benchmark for Training and Evaluating Ever-Evolving Language Models. In EMNLP 2022.

PDF Code

(2021). Towards Continual Knowledge Learning of Language Models. In ICLR 2022.

PDF Code

(2021). Spatial Dependency Parsing for Semi-Structured Document Information Extraction. In Findings of ACL 2021.

PDF

(2020). Is Retriever Merely an Approximator of Reader?. In Spa-NLP Workshop at ACL 2022.

PDF Code Poster Slides Video

(2020). ClovaCall: Korean Goal-Oriented Dialog Speech Corpus for Automatic Speech Recognition of Contact Centers. In INTERSPEECH 2020.

PDF Code

(2019). Efficient Dialogue State Tracking by Selectively Overwriting Memory. In ACL 2020.

PDF Code Video

(2019). Large-Scale Answerer in Questioner's Mind for Visual Dialog Question Generation. In ICLR 2019.

PDF Code Poster Slides