Hello! I'm Alessandro, currently pursuing my Bachelor's degree in Mathematical Engineering at Politecnico di Milano. My academic journey has led me to develop a strong interest in the field of Machine Learning, which I had the privilege to deepen during my time at Stanford University. My research currently focus on learning theory, particularly at the intersection between deep models and computational problems, or games. But not limited to it.
Meta-Learning Q-Function Priors for Efficient Few-Shot Imitation
We introduce a meta-learning approach that enables fast imitation learning from a few demonstrations, without manually designing rewards or relying on sample-inefficient inverse RL. By learning a Q-function prior across related tasks, our method quickly adapts to new tasks with just one gradient step, yielding both optimal policies and reward functions. Experiments on pixel-based tasks validate our approach.
Node Classification and Heuristic Search on the Rubik's Cube Graph with GNNs
This study focuses on the application of deep geometric models to solve the 3x3x3 Rubik's Cube. We begin by discussing the cube's graph representation and defining distance as the model's optimization objective. The distance approximation task is reformulated as a node classification problem, effectively addressed using Graph Neural Networks (GNNs).
[WIP] Approximating Combinatorial Optimization Algorithms with Generative Trajectory Modeling and Search
The paper tackles combinatorial problems with a generative model that provides solution trajectories, which doesn't rely on policy-gradient methods. By using a finite-horizon MDP as a compressed language for learning, the resulting supervised policy outperforms near-expert algorithms and remains competitive with state-of-the-art neural baselines.
Reti neurali artificiali
Si introduce il lettore alle reti neurali artificiali, una tecnologia quasi onnipervasiva nel campo dell'oggi tanto discusso apprendimento automatico. Caratterizzato da uno stile divulgativo ed un tono discorsivo, si rivolge soprattutto a chi non ha familiarità con aspetti tecnici.
