Check out our new preprint on the connections between machine learning and nonequilibrium physics: arxiv.org/abs/2306.03521. Shishir Adhikari started this as his last PhD work, and it has grown into a fun collaboration with Alkan Kabakcioglu, Alex Strang, and Deniz Yuret (1/n)

Incredibly proud of Ekin Akyürek for receiving the lexical semantics area award at ACL 2025. Come see his talk today at 4:15 in Pier 2/3! virtual2023.aclweb.org/paper_P2367.ht…

🎉 Exciting news! Our paper “CLIP-Guided StyleGAN Inversion for Text-Driven Real Image Editing" has been accepted to ACM Trans. on Graphics and will be presented at SIGGRAPH Asia ➡️ Hong Kong 2023. Co-authored with Canberk Baykal, Abdul Basit Anees, Duygu Ceylan, Erkut Erdem and Deniz Yuret. 1/4

Congratulations to Emirhan Kurtuluş for this amazing work!

Joint work with Barış Batuhan Topal (Barış Batuhan Topal), Prof. Deniz Yuret (Deniz Yuret). arxiv.org/abs/2211.10641

📢✨Next Tue (Nov 14) at 10:00 am, we'll have Kyunghyun Cho ( Kyunghyun Cho ) *in person* at KUIS AI Center, Koç University: "Beyond Test Accuracies for Studying Deep Neural Networks" for registration and details: [email protected] or just DM! #kuisaitalks #ArtificialInteligence

🎉 We are excited to announce the release of Kanarya 2B and Kanarya 0.7B, the latest pre-trained Turkish language models. 🎉

youtu.be/p5xWWrS8XLA?si… Yapay zekanın geçmişi, bugünü ve geleceği Prof. Dr. Deniz Yüret Koç Üniversitesi Öğretim Üyesi | İş Bank.Yapay Zeka Uygulama ve Araştırma Merkezi Müd. studioberlin Koç Üniversitesi #yapayzeka #DeepLearning #DenizYüret Deniz Yuret KUIS AI @kuisaicenter

1. 🧵🎉 Excited to share that our paper "Sequential Compositional Generalization in Multimodal Models" is accepted as a long paper at #NAACL2024! 🌟 We'll be presenting our findings in Mexico City this June (NAACL HLT 2025). Dive into the full details here 👇 - Paper:

🎉 Excited to share our new work: “Bridging the Bosphorus: Advancing Turkish Large Language Models through Strategies for Low-Resource Language Adaptation and Benchmarking”! #AI #NLProc #TurkishNLP 🇹🇷 🚀 📄Paper: arxiv.org/abs/2405.04685 🌐Website: emrecanacikgoz.github.io/Bridging-the-B… (1/7)

Türkiye Yapay Zeka İnisiyatifi’nin düzenlediği 7. TRAI Mayıs Çalıştayı’nda çok keyifli bir panelde buluştuk. Deniz Yuret Sabri Gökmenler Halil Aksu Türkiye Yapay Zeka İnisiyatifi #yapayzeka #AI

Projemizde ücretli görevlendirmek üzere STEM alanında eğitim görmüş kişiler arıyoruz! Çevrimiçi etiketleme yaparak saatlik 300 TL, toplamda 3000 TL’ye kadar Migros hediye çeki kazanabilirsiniz. İlgilenebilecek kişilerle paylaşabilir misiniz?

Deniz Yuret: Dördüncü Yapay Zeka Devrimi youtu.be/bU7oZRKBn60?si… via YouTube

Have you ever seen a learning curve that looks like a step function? It turns out a few hundred negative examples flips a switch inside an LLM and gives a discrete jump in accuracy. "How much do LLMs learn from negative examples?" (arxiv.org/abs/2503.14391) with Shadi Hamdan.

Training on wrong answers outpaces training on correct ones. 10 times more learning emerges from plausible errors than from truths. Large language models refine their accuracy slowly when they learn only from correct examples. This paper introduces Likra, which trains one

This plot compares how accuracy on the ARC-Challenge benchmark improves as more examples are used for two training methods. The blue line shows supervised fine-tuning using only correct question-answer pairs. Its accuracy rises slowly from about 60% with very few examples to