no code implementations • 5 May 2023 • Carlo Baldassi, Fabio Maccheroni, Massimo Marinacci, Marco Pirazzini
We develop a full-fledged analysis of an algorithmic decision process that, in a multialternative choice problem, produces computable choice probabilities and expected decision times.
no code implementations • 26 Apr 2023 • Carlo Baldassi, Enrico M. Malatesta, Gabriele Perugini, Riccardo Zecchina
We analyze the geometry of the landscape of solutions in both models and find important similarities and differences.
1 code implementation • 8 Feb 2022 • Carlo Baldassi
We present a meta-method for initializing (seeding) the $k$-means clustering algorithm called PNN-smoothing.
no code implementations • 7 Feb 2022 • Fabrizio Pittorino, Antonio Ferraro, Gabriele Perugini, Christoph Feinauer, Carlo Baldassi, Riccardo Zecchina
This lets us derive a meaningful notion of the flatness of minimizers and of the geodesic paths connecting them.
no code implementations • 19 Dec 2021 • Pietro Torta, Glen B. Mbeng, Carlo Baldassi, Riccardo Zecchina, Giuseppe E. Santoro
We apply digitized Quantum Annealing (QA) and Quantum Approximate Optimization Algorithm (QAOA) to a paradigmatic task of supervised learning in artificial neural networks: the optimization of synaptic weights for the binary perceptron.
no code implementations • 1 Oct 2021 • Carlo Baldassi, Clarissa Lauditi, Enrico M. Malatesta, Rosalba Pacelli, Gabriele Perugini, Riccardo Zecchina
Current deep neural networks are highly overparameterized (up to billions of connection weights) and nonlinear.
no code implementations • 2 Jul 2021 • Carlo Baldassi, Clarissa Lauditi, Enrico M. Malatesta, Gabriele Perugini, Riccardo Zecchina
The success of deep learning has revealed the application potential of neural networks across the sciences and opened up fundamental theoretical problems.
no code implementations • 27 Oct 2020 • Carlo Baldassi, Enrico M. Malatesta, Matteo Negri, Riccardo Zecchina
We analyze the connection between minimizers with good generalizing properties and high local entropy regions of a threshold-linear classifier in Gaussian mixtures with the mean squared error loss function.
no code implementations • 1 Aug 2020 • Carlo Baldassi, Fabio Maccheroni, Massimo Marinacci, Marco Pirazzini
Simulated Annealing is the crowning glory of Markov Chain Monte Carlo Methods for the solution of NP-hard optimization problems in which the cost function is known.
1 code implementation • ICLR 2021 • Fabrizio Pittorino, Carlo Lucibello, Christoph Feinauer, Gabriele Perugini, Carlo Baldassi, Elizaveta Demyanenko, Riccardo Zecchina
The properties of flat minima in the empirical risk landscape of neural networks have been debated for some time.
no code implementations • 3 May 2020 • Carlo Baldassi, Simone Cerreia-Vioglio, Fabio Maccheroni, Massimo Marinacci, Marco Pirazzini
We introduce an algorithmic decision process for multialternative choice that combines binary comparisons and Markovian exploration.
no code implementations • 15 Nov 2019 • Carlo Baldassi, Riccardo Della Vecchia, Carlo Lucibello, Riccardo Zecchina
The geometrical features of the (non-convex) loss landscape of neural network models are crucial in ensuring successful optimization and, most importantly, the capability to generalize well.
no code implementations • 29 Sep 2019 • Matteo Negri, Davide Bergamini, Carlo Baldassi, Riccardo Zecchina, Christoph Feinauer
Generative processes in biology and other fields often produce data that can be regarded as resulting from a composition of basic features.
no code implementations • 17 Jul 2019 • Carlo Baldassi, Enrico M. Malatesta, Riccardo Zecchina
Rectified Linear Units (ReLU) have become the main model for the neural units in current deep learning systems.
no code implementations • 20 May 2019 • Carlo Baldassi, Fabrizio Pittorino, Riccardo Zecchina
In the case of SGD and cross-entropy loss, we show that a slow reduction of the norm of the weights along the learning process also leads to WFM.
1 code implementation • 1 May 2019 • Carlo Baldassi
We compare this scheme with state-of-the-art alternative, a more standard genetic algorithm with deterministic pairwise-nearest-neighbor crossover and an elitist selection policy, of which we also provide an augmented and efficient implementation.
no code implementations • 26 Oct 2017 • Carlo Baldassi, Federica Gerace, Hilbert J. Kappen, Carlo Lucibello, Luca Saglietti, Enzo Tartaglione, Riccardo Zecchina
Stochasticity and limited precision of synaptic weights in neural network models are key aspects of both biological and hardware modeling of learning processes.
no code implementations • 3 Jul 2017 • Pratik Chaudhari, Carlo Baldassi, Riccardo Zecchina, Stefano Soatto, Ameet Talwalkar, Adam Oberman
We propose a new algorithm called Parle for parallel training of deep networks that converges 2-4x faster than a data-parallel implementation of SGD, while achieving significantly improved error rates that are nearly state-of-the-art on several benchmarks including CIFAR-10 and CIFAR-100, without introducing any additional hyper-parameters.
no code implementations • 26 Jun 2017 • Carlo Baldassi, Riccardo Zecchina
Their energy landscapes is dominated by local minima that cause exponential slow down of classical thermal annealers while simulated quantum annealing converges efficiently to rare dense regions of optimal solutions.
2 code implementations • 6 Nov 2016 • Pratik Chaudhari, Anna Choromanska, Stefano Soatto, Yann Lecun, Carlo Baldassi, Christian Borgs, Jennifer Chayes, Levent Sagun, Riccardo Zecchina
This paper proposes a new optimization algorithm called Entropy-SGD for training deep neural networks that is motivated by the local geometry of the energy landscape.
3 code implementations • 21 Aug 2016 • Carlo Baldassi
Our code for the Ising case is publicly available [https://github. com/carlobaldassi/RRRMC. jl], and extensible to user-defined models: it provides efficient implementations of standard Metropolis, the RRR method, the BKL method (extended to the case of continuous energy specra), and the waiting time method [Dall and Sibani Comput. Phys. Commun.
Statistical Mechanics Disordered Systems and Neural Networks
no code implementations • 20 May 2016 • Carlo Baldassi, Christian Borgs, Jennifer Chayes, Alessandro Ingrosso, Carlo Lucibello, Luca Saglietti, Riccardo Zecchina
We define a novel measure, which we call the "robust ensemble" (RE), which suppresses trapping by isolated configurations and amplifies the role of these dense regions.
no code implementations • 12 Feb 2016 • Carlo Baldassi, Federica Gerace, Carlo Lucibello, Luca Saglietti, Riccardo Zecchina
Learning in neural networks poses peculiar challenges when using discretized rather then continuous synaptic states.
no code implementations • 18 Nov 2015 • Carlo Baldassi, Alessandro Ingrosso, Carlo Lucibello, Luca Saglietti, Riccardo Zecchina
We introduce a novel Entropy-driven Monte Carlo (EdMC) strategy to efficiently sample solutions of random Constraint Satisfaction Problems (CSPs).
no code implementations • 18 Sep 2015 • Carlo Baldassi, Alessandro Ingrosso, Carlo Lucibello, Luca Saglietti, Riccardo Zecchina
We also show that the dense regions are surprisingly accessible by simple learning protocols, and that these synaptic configurations are robust to perturbations and generalize better than typical solutions.
no code implementations • 20 May 2015 • Carlo Baldassi, Alfredo Braunstein
The algorithm we present performs as well as BP on binary perceptron learning problems, and may be better suited to address the problem on fully-connected two-layer networks, since inherent symmetries in two layer networks are naturally broken using the MS approach.