Michela Proietti

[Michela Proietti @MProietti_]

6/

💡 Code available at: https://t.co/f87o6dO8D6

If you are interested, we are happy to engage in discussion and exchange ideas with others interested in brain–LLM alignment and interpretability!

[Michela Proietti @MProietti_]

We are excited to share our preprint presenting the first input attribution approach to investigate brain-LLM alignment: “Fine-Grained Analysis of Brain–LLM Alignment through Input Attribution”  🧠📚 with @webrot and @mtoneva1.

📄 Read the full paper: https://t.co/NKY0xLv9lv

🧵

[Michela Proietti @MProietti_]

5/

🧰 General framework: Beyond this study, our attribution method can help investigate how alignment arises by considering different models (e.g., trained with instruction-following objectives, brain-tuned, etc.), datasets (e.g. listening), or language tasks.

[Michela Proietti @MProietti_]

2️⃣Reviews of the state of the art in XAI-guided continual learning, highlighting the used benchmarks and learning scenarios.
3️⃣ Outlines future research directions, from self-interpretable architectures to new application domains and neuroscience-inspired models.

[Michela Proietti @MProietti_]

🚀 Excited to share: “XAI-Guided Continual Learning: Rationale, Methods, and Future Directions”, w/ @spideralessio @webrot.

📖 https://t.co/3MbhKZvnkV

💬 We’d be happy to chat more about this exciting direction! Feel free to reach out!

More info in 🧵.

@XAI_Research

[Michela Proietti @MProietti_]

💡 The paper:

1️⃣ Explores connections between artificial and biological neural networks that motivate the use of XAI in continual learning.

[Biagio La Rosa @larosabiagio]

Please retweet! We extended the deadline for our #XAI workshop XAI4DRL@AAAI2024 to Nov 21st! https://t.co/MpOtIt3qj8 Any XAI paper is welcome, even if RL is not involved! Please check the CFP and FAQ to know more about it!

[Biagio La Rosa @larosabiagio]

Please RT! Two weeks left to submit your work to our workshop on #eXplainableAI approaches for Deep #ReinforcementLearning #AAAI24, Vancouver (https://t.co/a5F8Krczpg) Deadline: Nov 15!