Vasilis Papageorgiou

Welcome! I am currently a third year PhD student at the Computer Sciences Department of UW-Madison, where I am fortunate enough to be working with Prof. Dimitris Papailiopoulos. My current research interests revolve around Large Language Models, and both theoretical and practical aspects of Machine Learning and Deep Learning.

Before coming to Madison, I earned my BSc and integrated MEng from the ECE department of the Technical University of Crete. During this time, I had the opportunity to work with Prof. Aggelos Bletsas, focusing on asynchronous inference algorithms for ambiently powered wireless sensor networks [Diploma Thesis].

news

Jul 23, 2023	I was invited to give a talk on LLMs at the 2023 ECE TUC Summer School on Data Analysis, hosted by my alma mater. [YouTube link]

selected publications

Preprint
From Artificial Needles to Real Haystacks: Improving Retrieval Capabilities in LLMs by Finetuning on Synthetic Data

Zheyang Xiong, Vasilis Papageorgiou, Kangwook Lee, and Dimitris Papailiopoulos

arXiv preprint arXiv:2406.19292, 2024

Abs arXiv Bib HTML

Recent studies have shown that Large Language Models (LLMs) struggle to accurately retrieve information and maintain reasoning capabilities when processing long-context inputs. To address these limitations, we propose a finetuning approach utilizing a carefully designed synthetic dataset comprising numerical key-value retrieval tasks. Our experiments on models like GPT-3.5 Turbo and Mistral 7B demonstrate that finetuning LLMs on this dataset significantly improves LLMs’ information retrieval and reasoning capabilities in longer-context settings. We present an analysis of the finetuned models, illustrating the transfer of skills from synthetic to real task evaluations (e.g., 10.5% improvement on 20 documents MDQA at position 10 for GPT-3.5 Turbo). We also find that finetuned LLMs’ performance on general benchmarks remains almost constant while LLMs finetuned on other baseline long-context augmentation data can encourage hallucination (e.g., on TriviaQA, Mistral 7B finetuned on our synthetic data cause no performance drop while other baseline data can cause a drop that ranges from 2.33% to 6.19%). Our study highlights the potential of finetuning on synthetic data for improving the performance of LLMs on longer-context tasks.
@article{xiong2024artificialneedlesrealhaystacks, title = {From Artificial Needles to Real Haystacks: Improving Retrieval Capabilities in LLMs by Finetuning on Synthetic Data}, author = {Xiong, Zheyang and Papageorgiou, Vasilis and Lee, Kangwook and Papailiopoulos, Dimitris}, journal = {arXiv preprint arXiv:2406.19292}, year = {2024}, }
Preprint
Everything Everywhere All at Once: LLMs can In-Context Learn Multiple Tasks in Superposition

Zheyang Xiong, Ziyang Cai , John Cooper , Albert Ge, Vasilis Papageorgiou, Zack Sifakis , Angeliki Giannou, Ziqian Lin, Liu Yang, Saurabh Agarwal, and others

arXiv preprint arXiv:2410.05603, 2024

Abs arXiv Bib HTML

Large Language Models (LLMs) have demonstrated remarkable in-context learning (ICL) capabilities. In this study, we explore a surprising phenomenon related to ICL: LLMs can perform multiple, computationally distinct ICL tasks simultaneously, during a single inference call, a capability we term "task superposition". We provide empirical evidence of this phenomenon across various LLM families and scales and show that this phenomenon emerges even if we train the model to in-context learn one task at a time. We offer theoretical explanations that this capability is well within the expressive power of transformers. We also explore how LLMs internally compose task vectors during superposition. Furthermore, we show that larger models can solve more ICL tasks in parallel, and better calibrate their output distribution. Our findings offer insights into the latent capabilities of LLMs, further substantiate the perspective of "LLMs as superposition of simulators", and raise questions about the mechanisms enabling simultaneous task execution.
@article{xiong2024everything, title = {Everything Everywhere All at Once: LLMs can In-Context Learn Multiple Tasks in Superposition}, author = {Xiong, Zheyang and Cai, Ziyang and Cooper, John and Ge, Albert and Papageorgiou, Vasilis and Sifakis, Zack and Giannou, Angeliki and Lin, Ziqian and Yang, Liu and Agarwal, Saurabh and others}, journal = {arXiv preprint arXiv:2410.05603}, year = {2024}, }