Program

Efficient Deep Learning: Methods and Applications

Workshop @ ICANN 2026

35th International Conference on Artificial Neural Networks

 Monday, September 14, 2026 - Padova, Italy

Room: Congress Centre B

Venue: Edificio PSICOLOGIA 3 – Scuola di Psicologia, Via Venezia, 14, Padova, Italy (map)

Session 1 - Introduction & Keynote

Session Chair:  Luca Pedrelli (Scuola Superiore Sant’Anna, Pisa, Italy)

9:30 Introduction

9:35 Efficient Deep Learning: Methods and Applications
Luca Pedrelli, Federico Aromolo, Stefano Dettori, Marco Cococcioni

10:00 Keynote speech
Is Efficiency Enough for Real-Time AI?
Giorgio Buttazzo (Scuola Superiore Sant’Anna, Pisa, Italy)

11:00 Coffee Break (Lunch and Poster Area)

Session 2 - Efficient RNNs and RL

Session Chair:  Marco Cococcioni (University of Pisa, Italy)

11:30 Deep Echo State Networks for Green Hydrogen Costs and Solar Energy Production Forecasting
Luca Pedrelli, Ismael Matino, Stefano Dettori, Valentina Colla

11:50 Dissecting Linear Recurrent Models: How Different Gating Strategies Drive Selectivity and Generalization
Younes Bouhadjar, Maxime Fabre, Felix Schmidt, Emre Neftci

12:10 Loss-Gated Planning: Improving Efficiency of Model-Based Reinforcement Learning
Lorenzo Vannini, Stefano Dettori, Luca Pedrelli, Giorgio Buttazzo, Valentina Colla

12:30 Interpretable Policy Distillation for Power Grid Topology Control
Aleksandra Dmitruka, Kārlis Freivalds

13:00 Lunch (Lunch And Poster Area)

Session 3 - Efficient Training and Inference

Session Chair:  Stefano Dettori (Scuola Superiore Sant’Anna, Pisa, Italy)

14:30 Towards Real-Time PixOOD: Efficient Anomaly Segmentation for Autonomous Vehicles
Luca de Martino, Federico Aromolo, Federico Nesti, Giorgio Buttazzo

14:50 Diffusion-Based Synthesis and Analysis of Neural Network Weights for Non-Trivial Tasks
Mykhailo Ruzmetov, Peter Sinčák

15:10 Branchless Software Emulation of FP8_E5M2 ML-MUL via 64-bit SWAR
Simone Gallo, Marco Cococcioni

15:30 BASIQ: Bayesian Sensitivity-Informed Quantization
Dmitri Demler, Thea Aarrestad

16:00 Coffee Break (Lunch and Poster Area)

Session 4 - Efficient LLMs

Session Chair:  Federico Aromolo (Scuola Superiore Sant’Anna, Pisa, Italy)

16:30 Small Microscaling Floating Point Formats for GPT-2 Large Language Models
Marco Cococcioni, Dario Pagani, Federico Rossi

16:50 Training-Efficient Collaborative LLM Decoding Using Self-Supervised Token-Level Routers
Federico Putamorsi, Giuseppe Trimigno, Monica Mordonini, Gianfranco Lombardo

17:10 Jailbreak Detection in LLMs: Do Interpretability Lenses Reveal Safety Signals?
Giuseppe Trimigno, Gianfranco Lombardo

18:00 Welcome Reception (Lunch and Poster Area)