Research Scientist at Bloomberg AI. Working with table retrieval and conversational AI. Opinions are my own.
London, England
Joined April 2017
- Tweets168
- Following196
- Followers338
- Likes706
Shuo Zhang retweeted
Compression Beyond the Uncompressed: A Two-Stage Training Recipe for Soft Context Compression in RAG
@70LuonHfXrUWIz3 et al. train via imitation on correct RAG answers, then RL on RAG failures, surpassing uncompressed RAG at 16× compression.
📝arxiv.org/abs/2609.05152
Shuo Zhang retweeted
COPE, their new Mixture-of-Experts architecture meant to master the nuances of proactive strategy selection, enhances both recommendation accuracy & strategy optimization, thereby providing a robust framework to model the dynamism of conversational recommendation
#SIGIR2026
(3/3)
Shuo Zhang retweeted
The paper, co-authored with our #AI Engineering group's @imsure318 and @sheffieldNLP's Xi Wang (@wangxieric), investigates the optimal timing & selection of preference elicitation strategies to advance conversational recommendation systems
bloom.bg/4b3Doct
#SIGIR2026
(2/3)
Shuo Zhang retweeted
In the "Personalized Recommendation & Preference Modeling" track at #SIGIR2026 in Melbourne today (11 AM AEST), @SheffieldNLP's Feng Xia presents "When and How to Ask: Dynamic Preference Elicitation Strategies for Conversational Recommendation"
bloom.bg/4fdl2bs
(1/3)
Shuo Zhang retweeted
#AI researcher @imsure318 helped organize today's Workshop on Proactive Conversational Information Seeking with Large Language Models (#ProActLLM) that's co-located with #CIKM2025
bloom.bg/43mOW6D
#LLMs #GenAI #NLProc #InformationRetrieval #HCI #HumanComputerInteraction
Shuo Zhang retweeted
Our #AI Engineering group’s @imsure318 co-authored "Enhancing #RAG Efficiency with Adaptive Context Compression" with Shuyu Guo (Shandong University) & Zhaochun Ren (Leiden University); it's being presented at #EMNLP2025 today (13:00 CST)
bloom.bg/48XV74J
#GenAI
(1/2)
Shuo Zhang retweeted
Our #AI Engineering group’s @imsure318 co-authored "Enhancing #RAG Efficiency with Adaptive Context Compression" with Shuyu Guo (Shandong University) & Zhaochun Ren (Leiden University); it's being presented at #EMNLP2025 today (13:00 CST)
bloom.bg/3XahMU9
#GenAI
(1/2)
ALT Read the whole "Enhancing RAG Efficiency with Adaptive Context Compression" paper in Findings of the ACL: EMNLP 2025: https://aclanthology.org/2025.findings-emnlp.1307.pdf
Shuo Zhang retweeted
The paper introduces Adaptive Context Compression for #RAG (ACC-RAG), a way to teach the #AI system to “skim smartly” — similar to how people don’t read every single word in an article but focus on the most relevant parts
bloom.bg/3XahMU9
#EMNLP2025
(2/2)
ALT Read the whole "Enhancing RAG Efficiency with Adaptive Context Compression" paper in Findings of the ACL: EMNLP 2025: https://aclanthology.org/2025.findings-emnlp.1307.pdf
Shuo Zhang retweeted
📢 Call For Papers: 1st workshop on Proactive Conversational Information Seeking with Large Language Models @cikm2025
For more info visit: proactllm.github.io/
📅 When: November 14, 2025
📝 Deadline: August 31, 2025
#ProActLLM #cikm2025 #LLM #ProactiveAI #ConversationalAI
Shuo Zhang retweeted
#AI researcher @imsure318 is an organizer of today's 2nd Workshop on Financial Information Retrieval in the Era of #GenerativeAI. FinIR 2025 addresses the challenges of using #GenAI in finance through advanced #InformationRetrieval techniques.
bloom.bg/4lGqHrk
#SIGIR2025
Happenging in Meeting Room PALLADIO B now
#AI researcher @imsure318 is an organizer of today's 2nd Workshop on Financial Information Retrieval in the Era of #GenerativeAI. FinIR 2025 addresses the challenges of using #GenAI in finance through advanced #InformationRetrieval techniques.
bloom.bg/44vFgHW
#SIGIR2025
Shuo Zhang retweeted
At #SIGIR2025 today (10:30 AM CEST), @UvA_Amsterdam’s @MarieHendriksen & @mdr & Bloomberg's @imsure318, @ridhorei, @myahya & @edgarmeij present “Benchmark Granularity and Model Robustness for Image-Text Retrieval: A Reproducibility Study”
bloom.bg/4kzweyf
(1/3)
Shuo Zhang retweeted
At #SIGIR2025 today (10:30 AM CEST), @UvA_Amsterdam’s @MarieHendriksen & @mdr & Bloomberg's @imsure318, @ridhorei, @myahya & @edgarmeij present “Benchmark Granularity and Model Robustness for Image-Text Retrieval: A Reproducibility Study”
bloom.bg/4eNXdVK
(1/3)
ALT Read the whole "Benchmark Granularity and Model Robustness for Image-Text Retrieval: A Reproducibility Study" paper in the SIGIR '25 Proceedings: https://dl.acm.org/doi/10.1145/3726302.3730290
Shuo Zhang retweeted
The paper's authors explored how robust vision-language models in real-world #multimodalIR systems are to textual perturbations by systematically testing models like ALIGN & CLIP, revealing surprising sensitivity to word order & granularity in benchmarks
#SIGIR2025
(2/3)
Shuo Zhang retweeted
"Benchmark Granularity and Model Robustness for Image-Text Retrieval: A Reproducibility Study," by @MarieHendriksen & @mdr (@UvA_Amsterdam) & our #AI engineers @imsure318, @ridhorei, @myahya & @edgarmeij, can be read in the #SIGIR2025 Proceedings at bloom.bg/40fw5sL
(3/3)
Shuo Zhang retweeted
Researchers and engineers from @Bloomberg's #AI Engineering group co-authored 3 papers at #SIGIR2025 & helped co-organize the #FinIR workshop; learn more about their research & how their work will advance the state-of-the-art in #informationretrieval
bloom.bg/44wckje
(1/2)
Shuo Zhang retweeted
They'll be presenting their paper, "Enhancing #QuestionAnswering on Charts Through Effective Pre-training Tasks," during today's @BlackboxNLP poster sessions (11 AM & 3:30 PM EST). Their paper can be found in the #EMNLP2024 Proceedings: bloom.bg/3O36cG8
#AI #NLProc
