Investigador Postdoc per treballar en la Representació per actuar i planificar en Intel·ligència Artificial

Realització d'activitats cientificotècniques, per a l'execució de la línia de recerca Representation Learning for Acting and Planning in Artificial Intelligence (Aprenentatge de representació per actuar i planificar).

Aprenentatge de models dinàmics en forma lògica, aprenentatge d'estructures de subobjectius, de polítiques generals i estructures jeràrquiques d'actuació i planificació. 

El treball inclou teoria i experiments i antecedents en aprenentatge profund, aprenentatge de reforç i planificació. Es preveu que s’ha de realitzar aquestes tasques en un termini aproximat d’un any.

Deadline: 10 d'octubre de 2023

Més informació aquí

• Bases de la convocatòria, 21/09/2023
• Resolució, 28/09/2023

Post-doctoral research fellow to contribute to the implementation of the ERC StG SCALER

The applicant will work on a variety of work packages of the SCALER project, and specifically work on exploring the following directions:
- The development of new generalization guarantees based on the "online-to-PAC conversion" technique recently developed by Lugosi & Neu (2023).
- The extension of the above techniques to provide uncertainty quantification tools for potentially nonlinear models (such as deep neural networks).
- Apply the above results to address challenging exploration-exploitation dilemmas in large-scale sequential decision making.
- Extend previously existing "information-theoretic" analysis techniques for sequential decision making using convex analytic techniques, or the online-to-PAC machinery mentioned above.

The successful applicant needs to demonstrate prior experience in relevant areas of machine learning theory. Familiarity with PAC-Bayesian generalization bounds, online learning, information theory, and reinforcement learning theory (including bandit theory) is a big advantage. The applicant needs to hold a PhD in computer science, mathematics, statistics, or related areas, and have a demonstrated track record of publishing at top venues like NeurIPS, ICML, COLT, ALT, or AISTATS.

Deadline: Oct 04, 2023

More information here

• Bases de la convocatòria, del 27/09/2023
• Resolució, 28/09/2023 

Post-doctoral position in Reinforcement Learning and Computational Neuroscience

We are looking for strong post-doctoral candidates interested in joining the group of Theoretical and Cognitive Neuroscience of Ruben Moreno-Bote lab and the group of Artificial Intelligence and Machine Learning of Anders Jonsson.

We are looking for excellent post-doctoral candidates with Physics, Mathematics or Machine Learning background. The publication record of the candidate should be very strong in their respective discipline. 

The candidate will develop state of the art theory and models of RL inspired by computational neuroscience to understand natural behavior based on reward-free approaches. Related papers on the focus of the project are:
[2205.10316] Seeking entropy: complex behavior from intrinsic motivation to occupy action-state path space (arxiv.org)
[2302.01098] A general Markov decision process formalism for action-state entropy-regularized reward maximization (arxiv.org)

The candidate will benefit from the stimulating environment of Barcelona area in Theoretical and Systems Neuroscience and will have the opportunity of enjoying a lively city. TCN lab (https://www.upf.edu/web/tcn) 
forms part of a larger network of neuroscience labs in Barcelona. 

Financial support is provided by MdM and salary is competitive.

APPLICATION: A full CV in pdf format, a 1 page-long project in pdf format, and names and email addresses of 2 referees should be sent to the following address: [email protected]

Informal inquiries are welcome.  

DEADLINE: until the position is filled

STARTING DATE: any time between November 2023 and April 2024