Liang Wu

[Liang Wu @wuliang211]

Your LLM already reasons. Why train another model just to make decisions?

AnyJev keeps decision-making in the LLM you already run: a calibration layer plus a closed-form head.

New question? Write a prompt. Zero labels.
Have 100–300 labels? One closed-form solve, seconds on a CPU.
Better LLM? Swap it in.

📊 JevBench public set, 231 typed decisions, Qwen3-8B, zero labels:
• 0.727 accuracy (easy 1.000, hard 0.523)
• Every answer schema-valid, 57 ms median
• One temperature: ECE 0.246 → 0.079, accuracy unchanged

With labels, the head costs 0.68–0.84× a forward pass.
Serves on vLLM: an embed server plus a ~100 KB head.

Keep the reasoning in your LLM.
Calibrate the probabilities, not another model.

pip install anyjev
🔗https://t.co/5qBrRMAMx4

[Liang Wu @wuliang211]

I just published Shrinking the GPU Footprint of LLM Agents at Nokia: Three Recent Papers https://t.co/6ocJyKbjVm

[Liang Wu @wuliang211]

@South_River1 @suzan @andrew_trotman As an industrial researcher, I find recommendation to be one of the most compelling topics in SIGIR, given its critical role in real-world applications beyond academic interests. It resolves information overloading and it naturally is part of IR.

[Liang Wu @wuliang211]

Finally

[CNBC Now @CNBCnow]

Airbnb releases public IPO filing; seeks Nasdaq listing under ticker ABNB
https://t.co/EEy2xl86RM

[Liang Wu @wuliang211]

@yanjiefoo Thanks Yanjie. The paper is available here: https://t.co/zXo1GWLtbD

[Liang Wu @wuliang211]

We proposed a location-based personalization model that improved 3.9% GMV of Airbnb Experiences in the online a/b test last year - the most influential single experiment in improving GMV in 2019. We are excited to share that the work was accepted by SIGIR'2020.

[Liang Wu @wuliang211]

Link to the paper: https://t.co/zXo1GWLtbD

[Liang Wu @wuliang211]

Interviewed with @Springboard on machine learning and our work in #AirbnbExperiences

[Springboard @Springboard]

New Real Talk: Springboard's @siiyeah interviewed @Airbnb Machine Learning Engineer @wuliang211 about Airbnb Experiences, fake news detection, #deeplearning, and🦀experiences: https://t.co/hiSKs5GmBr

#machinelearning #DeepLearning

[Liang Wu @wuliang211]

@SkilletDoux My favorite place is Nhat Pho Restaurant in Mesa.

[Liang Wu @wuliang211]

@miha_jlo @AirbnbData Congratulations!

[Liang Wu @wuliang211]

@BrianDavison Lol this is so true. The feeling is 'finally I can fully focus on this' 😂

[Liang Wu @wuliang211]

Joint work with @fredmorstatter and @liuhuan https://t.co/V8DGtmidIV

[Springer Linguistics @SpringerLing]

SlangSD: building, expanding and using a sentiment dictionary of #slang words for short-text sentiment classification
https://t.co/kFfb6odYPF
#new in Language Resources and Evaluation
#linguistics

[Kamelia Aryafar @KAryafar]

Curious how to use transfer learned features with (poly)LDA? read @murium9's latest paper on a method for creating a bag of words representation of unrolled image vectors here: https://t.co/h6ybDk6X8Q #machinelearninng #deeperlearning #topicmodeling

[Liang Wu @wuliang211]

@miha_jlo Congratulations!

[Liang Wu @wuliang211]

A preprint is available here: https://t.co/Yggi4rNIj6 https://t.co/E86NnCqlkl

[Liang Wu @wuliang211]

My internship project was accepted at SIGIR'18, joint efforts with Dr. Diane Hu, Dr. Liangjie Hong @hongliangjie and Dr. Huan Liu @liuhuan . The paper will be released soon but my internship report is available here: https://t.co/bRnW2wofBR #SIGIR2018 #SIGIR18 #SIGIR https://t.co/rJpAz98cun

[Liang Wu @wuliang211]

My internship project was accepted at SIGIR'18, joint efforts with Dr. Diane Hu, Dr. Liangjie Hong @hongliangjie and Dr. Huan Liu @liuhuan . The paper will be released soon but my internship report is available here: https://t.co/bRnW2wofBR #SIGIR2018 #SIGIR18 #SIGIR https://t.co/rJpAz98cun

[Liangjie Hong @hongliangjie]

Our paper "Turning Clicks into Purchases: Revenue Optimization for Product Search in E-Commerce" (by our last summer intern Liang Wu from ASU) has been accepted into #SIGIR2018 as a full paper. Want to work on ML problems in E-commerce? Join us!

[Liang Wu @wuliang211]

@liuhuan https://t.co/6zHLgemU9i

[Association for Computing Machinery @TheOfficialACM]

Yesterday, "The Morning Paper" examined “Tracing Fake News Footprints: Characterizing Social Media Messages by How They Propagate," from WISDM '18.

https://t.co/W2RF9IK4nY

[Tanu Mitra @tanmit]

At the mining misinformation tutorial #ICDM2017. We found similar temporal differences in our recent CSCW paper https://t.co/QIIKwh7C5g

[Liang Wu @wuliang211]

One paper accepted at SDM 2018, Toward Relational Learning with Misinformation: https://t.co/4nfuH4IRmY #SDM18 #SDM2018

[Liang Wu @wuliang211]

#SIAMSDM17 Our talk about Relational Learning! #MakeRLPersonalizedAgain. @LiJundong