Davide Paglieri
I am a second-year PhD Student at UCL,
advised by Tim Rocktäschel and Jack Parker-Holder at UCL DARK Lab.
I was previously a Research Engineer at Bending
Spoons.
I obtained my MSc in Computer Science (AI & ML) from Imperial College London, graduating
with Distinction. In my Master's thesis, I explored Open-Ended Reinforcement
Learning
for Dynamic Robot Locomotion, advised by Antoine Cully.
Prior to that, I obtained a BSc in Computer Engineering at Politecnico di Torino, graduating with
110/110 cum Laude.
My research interests include Large Language Models, Reinforcement Learning,
Diffusion Models,
Open-Endedness, and generalist AI agents.
Contact: paglieridavide [at] gmail [dot] com
|
|
|
BALROG: Benchmarking Agentic LLM and VLM Reasoning On Games
Davide Paglieri*, Bartłomiej Cupiał*, Samuel Coward, Ulyana
Piterbarg,
Maciej Wolczyk, Akbir Khan, Eduardo Pignatelli, Łukasz Kuciński,Lerrel Pinto, Rob
Fergus, Jakob Nicolaus Foerster, Jack Parker-Holder, Tim Rocktäschel
Parker-Holder, Tim Rocktäschel
preprint, 2024
Benchmarking LLM and VLM agents capabilities on long-horizon game environments
such as NetHack
|
|
Outliers and Calibration Sets have Diminishing Effect
on Quantization of Modern LLMs
Davide Paglieri, Saurabh Dash, Jack
Parker-Holder, Tim Rocktäschel
ICML @ ES-FOMO-II, 2024
Study uncovering the effect of outliers and calibrations sets in quantization of
modern LLMs
|
<
|
Assessing the Zero-Shot Capabilities of LLMs
for Action Evaluation in RL
Eduardo Pignatelli,
Johan Ferret,
Tim Rocktaschel,
Edward Grefenstette,
Davide Paglieri
Samuel Coward,
Laura Toni,
arxiv, 2024
Evaluating Automated Reinforcement Learning Credit Assignment with LLMs
|
|
Multi-Agent Diagnostics for Robustness via Illuminated Diversity
Mikayel Samvelyan*, Davide Paglieri*, Minqi Jiang, Jack
Parker-Holder, Tim Rocktäschel
AAMAS, 2024 (oral)
Uncovering vulnerabilities in multi-agent systems with the power of
open-endedness.
|
I previously worked as a Research Engineer at Bending
Spoons where I researched, prototyped, and deployed deep
learning models on several of the company's apps,
Remini, Splice, Dawn
AI, focusing on diffusion generative models, image enhancement
and
artificial slow motion.
Whilst there, I conceptualized and led the development of Dawn AI, a mobile app
leveraging generative diffusion models to create AI art. Initially, the app allowed
users to generate artwork from text, sketches, or images, and later expanded to
include AI-generated avatars.
As the AI lead, I guided the app to achieve a top
ranking in the US App Store (and other regions) for three consecutive days.
Eventually, Dawn AI's features were integrated into Remini AI.
As a result of my efforts on Dawn AI, I had the opportunity to present and
give
a demo to Tim Cook, Apple's
CEO, while he was visiting our office in Milan.
|
|
|