Sign in

Shaoxiong Ji

@jis11.bsky.social
19 followers 23 following 23 posts

ELLIS Institute Finland & University of Turku

PostsRepliesMedia
Shaoxiong Ji @jis11.bsky.social · 01/10/2026
Can LLMs reuse memory learned by another model? 🧠 Mingyuan Li presents cross-model memory transfer (#NeurIPS2026) and MemoryATHENA at the TurkuNLP Seminar. 🗓 Wed Oct 7, 13:00 EEST 📍 1171 TU3, Maarintie 8, Aalto / Zoom 🔗 www.olaresearch.org/seminar/2026...
olaresearch.org
Seminar Details | TurkuNLP Research Seminar
TurkuNLP Research Seminar - From Portable Memory to Adaptive Memory: Building Reusable Memory Interfaces for Large Language Models by Mingyuan Li.
011
Shaoxiong Ji @jis11.bsky.social · 30/09/2026
One year ago today, I moved my research group to @ellisinstitute.fi and the University of Turku. I wrote a look back. www.olaresearch.org/jis/blog/yea...
olaresearch.org
Looking Back on 2025–2026: Research, Supervision and Community | Blogs | Omni Language AI Research
A look back at 2025–2026: papers at NeurIPS, COLM, EMNLP and ACL, the FINe-Health Foundry project, FineOPUS, the junior researchers I work with, and the community events we organized.
020
Shaoxiong Ji @jis11.bsky.social · 30/09/2026
Big thanks to the team: Mingyuan Li, Guangsheng Yu, and all co-authors! Check out more from our lab at olaresearch.org Works made in @ellisinstitute.fi and @turkunlp.bsky.social ✨ #AI #MachineLearning #LLM
olaresearch.org
Omni Language AI Research (OLAResearch)
Distilling the essence of human language into artificial intelligence.
010
Shaoxiong Ji @jis11.bsky.social · 30/09/2026
2️⃣ New Preprint out! MemoryAthena: Adaptive Routing over Latent and Generated Memories. We propose a novel routing approach that bridges the gap between latent and explicit memory spaces for better context utilization. 🚀 🔗 olaresearch.org/MemoryATHENA
olaresearch.org
MemoryATHENA: Adaptive Routing over Latent and Generated Memories
A reader-first memory interface that selectively admits generated memory around a direct Engram reference pathway.
100
Shaoxiong Ji @jis11.bsky.social · 30/09/2026
1️⃣ XMemTransfer is heading to #NeurIPS2026! 🎉 We introduce "Cross-Model Memory Transfer via Target-Side Reader Adaptation" to enable seamless knowledge sharing across different models. 🔗 olaresearch.org/XMemTransfer
olaresearch.org
Cross-Model Memory Transfer via Target-Side Reader Adaptation
We study cross-model frozen-memory transfer, showing that target-side readers can extract and align representations across model architectures and tokenizers.
102
Shaoxiong Ji @jis11.bsky.social · 30/09/2026
Excited to share two new updates on AI Memory from our lab! 🧠👇
110
Shaoxiong Ji @jis11.bsky.social · 23/08/2026
We're hiring! 🚀 ELLIS Institute Finland is looking for PhDs, Postdocs & Entrepreneurial Postdocs for Autumn 2026. Join us to build versatile AI systems spanning: 🧠 ML (Continual Learning, RL, Memory) 🗣️ NLP (LLMs, Agents, Multilingual) 🏥 Healthcare AI www.olaresearch.org/news/ellis-r...
olaresearch.org
Job Openings: ELLIS Institute Finland PhD & Postdoc Recruitment Autumn 2026 | News | Omni Language AI Research
Job openings for PhD and Postdoc researchers, and an Entrepreneurial Postdoc at ELLIS Institute Finland in Autumn 2026.
013
Shaoxiong Ji @jis11.bsky.social · 03/08/2026
Join us for the next TurkuNLP Research Seminar, where Md Mohsinul Kabir (University of Manchester) will explore "Cross-Cultural Reasoning in Large Language Models." Date: Monday, August 10, 2026 Time: 14:30 - 15:30 (UTC+3) Full details & Zoom Link: olaresearch.github.io/seminar/2026...
olaresearch.github.io
Seminar Details | TurkuNLP Research Seminar
TurkuNLP Research Seminar - Cross-Cultural Reasoning in Large Language Models by Md Mohsinul Kabir.
001
Reposted by Shaoxiong Ji
ELLIS @ellis.eu · 09/07/2026
🏹 Job alert: Doctoral & Postdoctoral Researchers in Foundation Health Models at the ELLIS Institute Finland 📍 Espoo / Turku 🇫🇮 ⏰ 17 August 🔗 bit.ly/3QMzVYP
olaresearch.org
Job Openings: Doctoral & Postdoctoral Researchers in Foundation Health Models | News | Omni Language AI Research
Job openings for Doctoral Researcher and Postdoctoral Researcher in Foundation Health Models at the University of Turku and ELLIS Institute Finland.
064
Reposted by Shaoxiong Ji
ELLIS Institute Finland @ellisinstitute.fi · 21/04/2026
ELLIS Institute Finland's big ambitions: building the world's first nationwide foundation model for healthcare, boosting RDI with AI, bringing top talents to Europe. Read more about our first year and next steps from director @samikaski.bsky.social : www.ellisinstitute.fi/ellis-instit... @ellis.eu
Hand holding open laptop on country road with wind turbines, ELLIS Institute Finland logo and text below: machine learning fundamentals that meet real-world challenges.
Image credit: Zac Wolff on Unsplash
142
Shaoxiong Ji @jis11.bsky.social · 24/07/2026
4/ Whether you need a deep dive into scalable graph architectures or a practical guide to connecting graphs with LLMs, check out the pre-prints: 📖 Graph2text or Graph2token: arxiv.org/abs/2501.01124 📖 Graph Learning: arxiv.org/abs/2507.05636 #GraphNeuralNetworks #LLMs
arxiv.org
Graph2text or Graph2token: A Perspective of Large Language Models for Graph Learning
Graphs are data structures used to represent irregular networks and are prevalent in numerous real-world applications. Previous methods directly model graph structures and achieve significant success....
020
Shaoxiong Ji @jis11.bsky.social · 24/07/2026
3/ Both surveys highlight that the future of graph learning lies in: 🤖 Graph Foundation Models (GFMs) for cross-domain zero-shot transfer. 💡 Knowledge-infused Prompting to adapt LLMs to graph structures. 🛡️ Responsible AI to guarantee privacy, robustness, and fairness.
110
Shaoxiong Ji @jis11.bsky.social · 24/07/2026
2/ Next, "Graph Learning" by Feng Xia et al. (arXiv:2507.05636) is a monumental 185-page survey. 📚It maps the field into 6 key dimensions: Scalable architectures, Temporal dynamics, Multimodal inputs, Generative graphs, Explainability, and Responsible AI (privacy & fairness).
110
Shaoxiong Ji @jis11.bsky.social · 24/07/2026
1/ First up, "Graph2text or Graph2token" focuses on the intersection of graphs & LLMs. 🤖 It categorizes how LLMs ingest graphs into two paradigms: serializing graph structures to text (Graph2text) or projecting GNN embeddings as tokens (Graph2token).
100
Shaoxiong Ji @jis11.bsky.social · 24/07/2026
How do we scale representation learning and connect graphs with LLMs? 🌐🧠 Two recent heavy-hitting surveys provide the complete roadmap for modern Graph Learning and LLM4Graph! 🧵👇
121
Shaoxiong Ji @jis11.bsky.social · 20/07/2026
These two works are collaborative research projects by @ellisinstitute.fi , University of Turku, University of Helsinki, LMU München, and the University of Edinburgh!
000
Shaoxiong Ji @jis11.bsky.social · 20/07/2026
Both projects, including corpora, model weights, code, and evaluations, are fully open-source! 🔓 📄 MaLA Corpus: www.olaresearch.org/MaLA/ 📄 EMMA-500 Suite: www.olaresearch.org/EMMA-500-Gen2/ #NLProc #LLMs #MultilingualNLP #AI #MachineTranslation #OpenSource
olaresearch.org
MaLA: A Corpus and Data Mix for Massive Language Adaptation of Large Language Models
Introducing the MaLA suite: a 74B-token corpus spanning 939 languages, a 136B pre-training mix, and the EMMA-500 model optimized for massively multilingual adaptation.
110
Shaoxiong Ji @jis11.bsky.social · 20/07/2026
3/ The findings? 📈 Incorporating parallel bilingual data during continual pre-training (CPT) enhances cross-lingual transfer and downstream task performance, especially in low-resource settings.
100
Shaoxiong Ji @jis11.bsky.social · 20/07/2026
2/ EMMA-500 Suite (ACL 2026 Findings) asks the key question: Does bilingual translation data help during CPT? 🗺️ MaLA Translation Corpus: parallel translation data for 2,500+ language pairs. 🚀 EMMA suite (Mono & Bi models): trained on up to 671B tokens.
100
Shaoxiong Ji @jis11.bsky.social · 20/07/2026
1/ MaLA Corpus (COLM 2026) builds the foundation: 📦 MaLA Corpus: 74B tokens across 939 languages. ⚖️ Diverse Data Mix: balances low-resource texts with code, literature, and science to avoid catastrophic forgetting. 🤖 EMMA-500: 7B-based model optimized for cross-lingual transfer.
100
Shaoxiong Ji @jis11.bsky.social · 20/07/2026
How do we scale the multilingual capability of LLMs to hundreds of languages? And does parallel translation data actually help? 🌍 We are excited to share our two recent papers on massive language adaptation: MaLA Corpus (COLM 2026) & EMMA-500 Suite trained on parallel corpus (ACL 2026 Findings)! 🧵👇
121
Shaoxiong Ji @jis11.bsky.social · 02/07/2026
📢 Hiring in Health AI! 🇫🇮 We're recruiting a PhD & Postdoc to build health foundation models at @UniTurku & @ellisinstitute.fi 📅 Deadline: Aug 17, 2026 Apply: www.olaresearch.org/news/job-ope... #AI #HealthAI #Hiring #Postdoc #PhD
olaresearch.org
Job Openings: Doctoral & Postdoctoral Researchers in Foundation Health Models | News | Omni Language AI Research
Job openings for Doctoral Researcher and Postdoctoral Researcher in Foundation Health Models at the University of Turku and ELLIS Institute Finland.
032
Reposted by Shaoxiong Ji
ELLIS Institute Finland @ellisinstitute.fi · 09/02/2026
Large language models that are useful beyond English: introducing @jis11.bsky.social, one of our new principal investigators and assistant professor at @utu.fi. Ji is interested in multimodal AI for medical use and developing efficient LLMs Read more: www.ellisinstitute.fi/meet-our-new... @ellis.eu
022
Shaoxiong Ji @jis11.bsky.social · 12/01/2026
Join us to advance ML and NLP!
010
Reposted by Shaoxiong Ji
ELLIS Institute Finland @ellisinstitute.fi · 12/01/2026
#Hiring: postdocs in artificial intelligence and machine learning research. Deadline February 9, 2026. - Cutting-edge computational resources like @lumi-supercomputer.eu - Close collaboration with @ellis.eu - Ambitious, high-impact projects across Finland ➡️ www.ellisinstitute.fi/postdoc-recr...
Text: We're hiring! Postdoctoral fellows in artificial intelligence and machine learning research.
Below: ELLIS Institute Finland logo
198
Shaoxiong Ji @jis11.bsky.social · 03/09/2025
The TurkuNLP research group at the University of Turku, Finland is seeking two fully-funded PhD Researchers to join our team! Check out the website haif.utu.fi for details
haif.utu.fi
HAIF – Human-Centric Artificial Intelligence for Sustainable Future
HAIF is co-funded by European Union’s Horizon Europe research and innovation programme’s Marie Skłodowska-Curie Action.
011
Reposted by Shaoxiong Ji
ELLIS Institute Finland @ellisinstitute.fi · 20/08/2025
Welcome new principal investigators! 9 leading scholars and rising stars in AI & machine learning are joining ELLIS Institute Finland, part of the @ellis.eu network. Read more about our new professors: ➡️ www.ellisinstitute.fi/PIs-2025
Nine people are shown in different settings, photos arranged in a grid.
1176
Shaoxiong Ji @jis11.bsky.social · 02/07/2025
I'll be moving to ELLIS Institute Finland and the University of Turku this fall and looking to hire 2 PhD researchers and 1 Postdoctoral Researcher to work on exciting topics related to Large Language Models, AI for Health, or Multimodal Learning.
ats.talentadore.com
A Postdoctoral Researcher and two Doctoral Researchers in AI and Foundation Models
The Department of Computing is looking for one Postdoctoral Researcher and two Doctoral Researchers to join a research group of ELLIS PI, Assistant Professor Shaoxiong Ji. You will be working alongsid...
020