PhD · Deep Reinforcement Learning · University of Padova

Alberto Sinigaglia

Researcher in Deep Reinforcement Learning, Neural Networks, and Trustworthy AI. Interested in adversarial robustness, Bayesian ML, and fairness.

CV LinkedIn GitHub Stack Overflow Google Scholar

About

From Veneto, Italy — based in Padova
PhD in Deep Reinforcement Learning, University of Padova (Oct 2023 – ongoing)
Master's in Data Science · Bachelor's in Computer Science — both 110/110 with honors
Teaching Reinforcement Learning at Master level; extra lectures in the PhD Deep Learning course
Research interests: Deep RL, Neural Networks, Trustworthy AI, Adversarial Learning, Bayesian ML, Fairness
Always building something — lately all ML

Experience

Quantitative Researcher Jun 2026 – Aug 2026

G-Research · London, Soho

Applied Scientist Researcher Nov 2025 – Apr 2026

Amazon · Luxembourg

LLMs: agentic integration of AWS pipelines using the MCP protocol
Graph Neural Networks: metaheuristic search using a pessimistic GNN ensemble
Deep RL: planning agent exploiting auto-regressive Q-policies

Professor Assistant Oct 2023 – ongoing

University of Padova

Co-teaching Reinforcement Learning at the Master in Control System Engineering
Extra lectures at the PhD course in Deep Learning, Dept. of Engineering

Awards

2019–2022 4× winner of scholarship "1000 e 1 lode" — awarded to the top 3% of students each year
2025 Web Chair · IFAC Joint Conference on Computers, Cognition and Communication (J3C)
2025 Associate Editor · IFAC Joint Conference on Computers, Cognition and Communication (J3C)

Publications

NeurIPS 2025

Simple and Effective Specialized Representations for Fair Classifiers

Sinigaglia, Alberto* · Sartor, Davide* · et al.

ICML 2025

Advancing Constrained Monotonic Neural Networks: Achieving Universal Approximation Beyond Bounded Activations

Sinigaglia, Alberto* · Sartor, Davide* · et al.

IEEE Trans. Automation Science & Engineering 2025

Edge Delayed Deep Deterministic Policy Gradient: Efficient Continuous Control for Edge Scenarios

Sinigaglia, Alberto

IEEE Robotics & Automation Magazine 2025

Reinforcement Learning for Robust Athletic Intelligence: Lessons from the 2nd AI Olympics with RealAIGym Competition

Wiebe, Felix · Sinigaglia, Alberto · et al.

ICONS 2025

Deep Reinforcement Learning for Autonomous Navigation: Sim-to-Real Transfer on TurtleBots

Dal Nevo, Matteo · Sinigaglia, Alberto · et al.

ICONS 2025

Exploiting Estimation Bias in Clipped Double Q-Learning for Continuous Control Reinforcement Learning Tasks

Turcato, Niccolò · Sinigaglia, Alberto · et al.

ICONS 2025

AI Olympics Challenge with Evolutionary Soft Actor Critic

Calì, Marco · Sinigaglia, Alberto · et al.

ICONS 2025

Multi-layer Abstraction for Nested Generation of Options (MANGO) in Hierarchical Reinforcement Learning

Arcudi, Alessio · Sinigaglia, Alberto · et al.

Information Sciences 2024

On the Limitations of Adversarial Training for Robust Image Classification with Convolutional Neural Networks

Carletti, Mattia · Sinigaglia, Alberto · et al.

IEEE Access 2024

Adversarially Robust Fault Zone Prediction in Smart Grids with Bayesian Neural Networks

Efatinasab, Emad · Sinigaglia, Alberto · et al.

Projects

RL Snake

Snake with rooms and walls, solved with DQN, A2C and PPO.

Force / Self-Organizing Graph

Force-directed graph layout via non-convex gradient optimization in JS.

Block Coordinates Gradient Descent

Comparison of BCGD methods on synthetic and real datasets.

Stochastic Gradient Descent Alternatives

Comparison of SARAH, SpiderBoost and SNVRG on real-world datasets.

Attention Sampling Classification

Mega-pixel image classification via attention sampling trained with RL — orders of magnitude fewer FLOPs than standard CNNs.

Binarized Neural Networks

Neural networks with binary weights, implemented with first-order logic and MaxSAT solvers.

Discrete GANs

GANs with a discrete generator trained via RL and straight-through estimator.

Bonus

You took your time to get to here, so here's a little reward
(some of my favourite animations that I've developed... not too mobile-friendly):