Source author record

Marco Romanelli

Marco Romanelli appears in the imported research catalog. Authorship, coauthor and topic links are available while profile ownership is still unclaimed.

ResearcherUnclaimed source record

physics.optics Computer Vision Information Theory Machine Learning math.IT nlin.AO nlin.CD

Catalog footprint

What is connected

5works

7topics

4close collaborators

Actions

Connect this record

Open graph Browse works

Inspect adjacent papers, topics, institutions and collaborators without losing the researcher page.

Building this map preview

BZPEER is loading the nearby papers, people, topics and institutions for this page.

preprint2022arXiv

MEAD: A Multi-Armed Approach for Evaluation of Adversarial Examples Detectors

Detection of adversarial examples has been a hot topic in the last years due to its importance for safely deploying machine learning algorithms in critical applications. However, the detection methods are generally validated by assuming a single implicitly known attack strategy, which does not necessarily account for real-life threats. Indeed, this can lead to an overoptimistic assessment of the detectors' performance and may induce some bias in the comparison between competing detection schemes. We propose a novel multi-armed framework, called MEAD, for evaluating detectors based on several attack strategies to overcome this limitation. Among them, we make use of three new objectives to generate attacks. The proposed performance metric is based on the worst-case scenario: detection is successful if and only if all different attacks are correctly recognized. Empirically, we show the effectiveness of our approach. Moreover, the poor performance obtained for state-of-the-art detectors opens a new exciting line of research.

preprint2020arXiv

Feature selection in machine learning: Rényi min-entropy vs Shannon entropy

Feature selection, in the context of machine learning, is the process of separating the highly predictive feature from those that might be irrelevant or redundant. Information theory has been recognized as a useful concept for this task, as the prediction power stems from the correlation, i.e., the mutual information, between features and labels. Many algorithms for feature selection in the literature have adopted the Shannon-entropy-based mutual information. In this paper, we explore the possibility of using Rényi min-entropy instead. In particular, we propose an algorithm based on a notion of conditional Rényi min-entropy that has been recently adopted in the field of security and privacy, and which is strictly related to the Bayes error. We prove that in general the two approaches are incomparable, in the sense that we show that we can construct datasets on which the Rényi-based algorithm performs better than the corresponding Shannon-based one, and datasets on which the situation is reversed. In practice, however, when considering datasets of real data, it seems that the Rényi-based algorithm tends to outperform the other one. We have effectuate several experiments on the BASEHOCK, SEMEION, and GISETTE datasets, and in all of them we have indeed observed that the Rényi-based algorithm gives better results.

preprint2016arXiv

Excitable-like chaotic pulses in the bounded-phase regime of an opto-radiofrequency oscillator

We report theoretical and experimental evidence of chaotic pulses with excitable-like properties in an opto-radiofrequency oscillator based on a self-injected dual-frequency laser. The chaotic attractor involved in the dynamics produces pulses that, albeit chaotic, are quite regular: They all have similar amplitudes, and are almost periodic in time. Thanks to these features, the system displays properties that are similar to those of excitable systems. In particular, the pulses exhibit a threshold-like response, of well-defined amplitude, to perturbations, and it appears possible to define a refractory time. At variance with excitability in injected lasers, here the excitable-like pulses are not accompanied by phase slips.

preprint2016arXiv

Frequency-locked chaotic opto-RF oscillator

A driven opto-RF oscillator, consisting of a dual-frequency laser (DFL) submitted to frequency-shifted feedback, is studied experimentally and numerically in a chaotic regime. Precise control of the reinjection strength and detuning permits to isolate a parameter region of bounded-phase chaos, where the opto-RF oscillator is frequency-locked to the master oscillator, in spite of chaotic phase and intensity oscillations. Robust experimental evidence of this synchronization regime is found and phase noise spectra allows to compare phase-locking and bounded-phase chaos regimes. In particular, it is found that the long-term phase stability of the master oscillator is well transferred to the opto-RF oscillator even in the chaotic regime.

preprint2014arXiv

Measuring the universal synchronization properties of coupled oscillators across the Hopf instability

When a driven oscillator loses phase-locking to a master oscillator via a Hopf bifurcation, it enters a bounded-phase regime in which its average frequency is still equal to the master frequency, but its phase displays temporal oscillations. Here we characterize these two synchronization regimes in a laser experiment, by measuring the spectrum of the phase fluctuations across the bifurcation. We find experimentally, and confirm numerically, that the low frequency phase noise of the driven oscillator is strongly suppressed in both regimes in the same way. Thus the long-term phase stability of the master oscillator is transferred to the driven one, even in the absence of phase-locking. The numerical study of a generic, minimal model suggests that such behavior is universal for any periodically driven oscillator near a Hopf bifurcation point.