Iñigo Alonso

Research Associate at @EdinburghNLP | Former
Ph.D. student at @Hitz_zentroa | NLP | Multimodality | Table Understanding

[Iñigo Alonso @alonsonlp]

Built alongside the wonderful people at @EdinburghNLP and @hitz_center

Read more about our paper here:
https://t.co/NN25u6bOit

[Iñigo Alonso @alonsonlp]

I’ll be presenting TABLET at #ICLR2026 this week! Our 4 million example Visual Table Understanding dataset with original table visualisations! Come and say hi! :D We’ll have actual Scottish tablets 👀

See you all on Friday at 7:15 PM Pavilion 4 P4-#3606

https://t.co/3IKS3rNOJt

[Irina Saparina @irisaparina]

Reasoning models are powerful, but they burn thousands of tokens on potentially wrong interpretations for ambiguous requests!

👉 We teach models to think about intent first and provide all interpretations and answers in a single response via RL with dual reward.

🧵1/6

[Ashutosh Adhikari @aadhikariii]

Excited to share my first work as a PhD student at @EdinburghNLP that I will be presenting at EMNLP!

RQ1: Can we achieve scalable oversight across modalities via debate?

Yes! We show that debating VLMs lead to better model quality of answers for reasoning tasks.

[HiTZ zentroa (EHU) @Hitz_zentroa]

Atzo eta gaur euskarazko ereduen artean onena aukeratzen utzi diegu #IEB2024|ko parte hartzaileei. Mila esker deneri!!

Sistema komertzialez gain HiTZ garatzen ari den Latxa 🐑berri txikiena ere probatu dugu, 8B tamainakoa. Llama 3.1-en oinarritzen da.

Eta irabazlea… 🥁

[Masaru Isonuma @m_isonuma]

New paper: "What's New in My Data? Novelty Exploration via Contrastive Generation," w/ @iatitov at @EdinburghNLP
https://t.co/vXghE4StN1

With no access to the data, but only access to the pre-trained and fine-tuned model, we can reveal novel aspects of the fine-tuning dataset.

[HiTZ zentroa (EHU) @Hitz_zentroa]

WE ARE HIRING!!
The HiTZ Center (https://t.co/pXKSIOPmm7) at the University of the Basque Country (UPV/EHU) invites applications for several funded research engineering and pre/postdoctoral positions in Natural Language and Speech Processing.
https://t.co/wDAZRzRjrL

[HiTZ zentroa (EHU) @Hitz_zentroa]

.@alonsonlp also participated in the panel about ArgNLE

[Iñigo Alonso @alonsonlp]

@agostina_cal @Wenhao_NLP and I went to Shanghai to see cool tall buildings 😎

[EdinburghNLP @EdinburghNLP]

Represent!!! 🚀🚀🚀🚀🚀

[Iñigo Alonso @alonsonlp]

@jparag123 Parag! I didn’t get to see you yet! We share the same poster session, but I’ll try to pass by yours :D

[Parag Jain ✈️ NeurIPS @jparag123]

Happening today at 11, In-Person Poster Session 1 😀

[Anar @anaryegen]

I will be presenting our work tomorrow at the Poster Session at 16:00 @ #ACL2024NLP
Feel free to come by and chat!

[Julen Etxaniz @juletxara]

Today I will be presenting Latxa at @aclmeeting In-Person Poster Session 3 at 16:00-17:30. Come if you want to hear about our work. I will be giving out Latxa stickers too! #ACL2024NLP

[Iñigo Alonso @alonsonlp]

Hey! I will be presenting our work, PixT3: Pixel-based Table-To-Text Generation, tomorrow at Poster Session 1 at 11:00AM during #ACL2024NLP Come and say hi! :D

[Iñigo Alonso @alonsonlp]

Reimagining table representation! In our new #ACL2024NLP paper we introduce PixT3: a family of image-based Table-to-Text Generation models that scale better at generating text from large tables, outperforming traditional text-based baselines.
https://t.co/wCylW9DdsE

[Iñigo Alonso @alonsonlp]

Conclusions: (i) Overall performance of LLMs with or without RAG still has large room for improvement; (ii) Performance for non-English languages substantially lower - stresses urgent need of advancing Medical QA in languages different to English

[Iñigo Alonso @alonsonlp]

New @hitz_zentroa paper! "MedExpQA: Multilingual Benchmarking of Large Language Models for Medical Question Answering"; w/ @ragerri and @jiporanm in Artificial Intelligence in Medicine (Elsevier)

Paper: https://t.co/iXH8ovdamD
Resources: https://t.co/pcBAF2FKfN

[Iñigo Alonso @alonsonlp]

Highlights: (i) MedExpQA: the first multilingual benchmark for MedicalQA including gold reference explanations; (ii) Exhaustive comparison of gold reference explanations with respect to automatically retrieved medical knowledge using state-of-the-art RAG techniques

[HiTZ zentroa (EHU) @Hitz_zentroa]

Come and work with us!

[Mustafa Shukor @MustafaShukor1]

LLMs are the de facto building block in recent multimodal models. Yet, it is still unclear why text-only models can generalize to multimodal inputs!

We investigate why, and provide insights with practical implications on performance, efficiency and safety problems. (1/10)