📣New paper!
We replace the entropy bonus in PPO with a complexity bonus, encouraging structured and stochastic policies that are robust to different hyperparameters and can work in envs with variable exploration needs.
Read more:
arxiv.org/abs/2509.20509
w/ @mircomusolesi
Two in a row! 🤯 Another paper out now on arXiv:
arxiv.org/abs/2502.14037
With @mircomusolesi, we develop a new, intuitive decoding strategy for language modeling based on math analysis, showing how it can foster diversity and accuracy of generated text.
New preprint out!🎉 Really happy to share one of the last (and most important) bits of my PhD thesis.
@mircomusolesi and I propose an information theoretic score to evaluate and optimize for value and originality in large language models. Read more here: arxiv.org/abs/2502.13207
Great time at the @genlawcenter workshop at #ICML2024! Super interesting talks and posters, plus the possibility of presenting our latest article on training foundation models as data compression and its copyright implications - already available on arXiv: arxiv.org/abs/2407.13493
Are you in Vienna for #ICML2024 and looking for a workshop on Saturday? Interested in how SSL can be seen as compressing training data and its impact on copyright? Join me and @mircomusolesi at GenLaw'24 for our spotlight presentation!
Read it on arXiv:
arxiv.org/abs/2407.13493
It's been a long journey, but our survey is now out in ACM Computing Surveys!
With @mircomusolesi, we review both generative and evaluative ML techniques for creativity, outlining the current research challenges and emerging opportunities.
dl.acm.org/doi/10.1145/36645…
New preprint available!
With @mircomusolesi, we investigate whether dream-like imagined trajectories can help RL agents generalize better when a limited amount of real experience is available:
arxiv.org/abs/2403.07979
Our latest paper is now out in the Journal of Artificial intelligence Research!
More than just RLHF, with @mircomusolesi we survey the growing field of RL for Generative AI, discussing opportunities and future challenges.
New Article: "Reinforcement Learning for Generative AI: State of the Art, Opportunities and Open Research Challenges" by Franceschelli and Musolesi jair.org/index.php/jair/arti…
Giorgio Franceschelli retweeted
Our paper with @Gionceschelli "Reinforcement Learning for Generative AI: State of the Art, Opportunities and Open Research Challenges" has been accepted in the Journal of Artificial Intelligence Research (JAIR). You can find the preprint version here: arxiv.org/abs/2308.00031
Giorgio Franceschelli retweeted
📖For our weekly @MilaNLProc lab seminar, it was a pleasure to have @Gionceschelli for a talk about Artificial Creativity.
#NLProc
LLMs are doing incredible things, but can we actually call them creative? "On the Creativity of Large Language Models" with @mircomusolesi discusses this and more, and you can find it also among @mtlaiethics briefs:
montrealethics.ai/on-the-cre…
Giorgio Franceschelli retweeted
You might be interested in this pre-print with @Gionceschelli on RL for AI, which came out in the middle of the summer holiday season: "Reinforcement Learning for Generative AI: State of the Art, Opportunities and Open Research Challenges".
arxiv.org/abs/2308.00031
Giorgio Franceschelli retweeted
New preprint with @Gionceschelli: "Reinforcement Learning for Generative AI: State of the Art, Opportunities and Open Research Challenges".
arxiv.org/abs/2308.00031
Giorgio Franceschelli retweeted
Article on @TheOfficialACM News by @purplesasa on measuring AI creativity, also featuring our work in this area.
cacm.acm.org/news/275135-mea…
\w @Gionceschelli
Giorgio Franceschelli retweeted
Article in @TheAtlantic on the creativity of Large Language Models, also citing our recent paper on the topic, with a quote by @Gionceschelli. theatlantic.com/technology/a…
Giorgio Franceschelli retweeted
New preprint with @Gionceschelli: "On The Creativity of Large Language Models".
arxiv.org/abs/2304.00008
1/5
Giorgio Franceschelli retweeted
Greatly expanded version of our review paper with @Gionceschelli "Creativity and Machine Learning: A Survey" now available on arXiv (lots happened in the area of generative learning in the past year).
Link: arxiv.org/pdf/2104.02726.pdf
Giorgio Franceschelli retweeted
Copyright in generative deep learning
@Gionceschelli and @MircoMusolesi
→ doi.org/10.1017/dap.2022.10
#DeepLearning #IntellectualProperty #IP #AI #MachineGeneratedArt #Artworks #Copyright #GenerativeDeepLearning #AIArt #AIArtworks #GenerativeArt #CopyrightLaw #CodeGeneration #ML
Giorgio Franceschelli retweeted
Our paper "Copyright in Generative Deep Learning" with @Gionceschelli has been accepted in @Data_and_Policy. We added an entire new section on code generation to the revised version.
Preprint version available here: arxiv.org/abs/2105.09266
Giorgio Franceschelli retweeted
New preprint with @Gionceschelli: "DeepCreativity: Measuring Creativity with Deep Learning Techniques" arxiv.org/abs/2201.06118















