A5028893275- Muscle activation and electromyography studies
- Action Observation and Synchronization
- Motor Control and Adaptation
- Reinforcement Learning in Robotics
- Sports Performance and Training
- Neural Networks and Applications
- Model Reduction and Neural Networks
- Auction Theory and Applications
- EEG and Brain-Computer Interfaces
- Neural dynamics and brain function
- Markov Chains and Monte Carlo Methods
- Advanced Numerical Methods in Computational Mathematics
- Mathematical Biology Tumor Growth
- Advanced Numerical Analysis Techniques
- Consumer Market Behavior and Pricing
- Advanced Bandit Algorithms Research
- Stochastic Gradient Optimization Techniques
- Children's Physical and Motor Development
- Medical Image Segmentation Techniques
- Lattice Boltzmann Simulation Studies
- Explainable Artificial Intelligence (XAI)
École Polytechnique Fédérale de Lausanne
2023-2024
Politecnico di Milano
2019
Proprioception tells the brain state of body based on distributed sensory neurons. Yet, principles that govern proprioceptive processing are poorly understood. Here, we employ a task-driven modeling approach to investigate neural code neurons in cuneate nucleus (CN) and somatosensory cortex area 2 (S1). We simulated muscle spindle signals through musculoskeletal generated large-scale movement repertoire train networks 16 hypotheses, each representing different computational goals. found...
Effective budget allocation is crucial for optimizing the performance of digital advertising campaigns. However, development practical algorithms remain limited, primarily due to lack public datasets and comprehensive simulation environments capable verifying intricacies real-world advertising. While multi-armed bandit (MAB) have been extensively studied, their efficacy diminishes in non-stationary where quick adaptation changing market dynamics essential. In this paper, we advance field by...
Efficient musculoskeletal simulators and powerful learning algorithms provide computational tools to tackle the grand challenge of understanding biological motor control. Our winning solution for inaugural NeurIPS MyoChallenge leverages an approach mirroring human skill learning. Using a novel curriculum approach, we trained recurrent neural network control realistic model hand with 39 muscles rotate two Baoding balls in palm hand. In agreement data from subjects, policy uncovers small...
In Reinforcement Learning, agents learn policies by exploring and interacting with the environment. Due to curse of dimensionality, learning that map high-dimensional sensory input motor output is particularly challenging. During training, state art methods (SAC, PPO, etc.) explore environment perturbing actuation independent Gaussian noise. While this unstructured exploration has proven successful in numerous tasks, it can be suboptimal for overactuated systems. When multiple actuators,...
Proprioception tells the brain state of body based on distributed sensors in body. However, principles that govern proprioceptive processing from those are poorly understood. Here, we employ a task-driven neural network modeling approach to investigate code neurons both cuneate nucleus (CN) and somatosensory cortex area 2 (S1). We simulated muscle spindle signals through musculoskeletal generated large-scale, naturalistic movement repertoire train thousands models 16 behavioral tasks, each...
Biological and artificial agents need to deal with constant changes in the real world. We study this problem four classical continuous control environments, augmented morphological perturbations. Learning locomote when length thickness of different body parts vary is challenging, as policy required adapt morphology successfully balance advance agent. show that a based on proprioceptive state performs poorly highly variable configurations, while an (oracle) agent access learned encoding...
Online advertising has become one of the most successful business models internet era. Impression opportunities are typically allocated through real-time auctions, where advertisers bid to secure advertisement slots. Deciding best for an impression opportunity is challenging, due stochastic nature user behavior and variability traffic over time. In this work, we propose a framework training auto-bidding agents in multi-slot second-price auctions maximize acquisitions (e.g., clicks,...
We present a deep reinforcement learning approach to classical problem in fluid dynamics, i.e., the reduction of drag bluff body. cast as discrete-time control with continuous action space: at each time step, an autonomous agent can set flow rate two jets fluid, positioned back The agent, trained Proximal Policy Optimization, learns effective strategy make interact vortexes wake, thus reducing drag. To tackle computational complexity dynamics simulations, which would training procedure...