Papers by Alessandro Ghio
Local Rademacher Complexity: Sharper risk bounds with and without unlabeled samples
Neural Networks, 2015
We derive in this paper a new Local Rademacher Complexity risk bound on the generalization abilit... more We derive in this paper a new Local Rademacher Complexity risk bound on the generalization ability of a model, which is able to take advantage of the availability of unlabeled samples. Moreover, this new bound improves state-of-the-art results even when no unlabeled samples are available.
Machine learning approaches for improving condition-based maintenance of naval propulsion plants
P. Sobolewski, M. Woźniak
Byte The Bullet: Learning on Real-World Computing Architectures
A. Danylenko, J. Lundberg, W. Löwe
Learning with Few Bits on Small–Scale Devices: from Regularization to Energy Efficiency
Human activity recognition on smartphones for mobile context awareness
A Heuristic Approach to Model Selection for Online Support Vector Machines
A learning machine with a bit-based hypothesis space
Human activity and motion disorder recognition: Towards smarter interactive cognitive environments
Energy Efficient Smartphone-Based Activity Recognition using Fixed-Point Arithmetic
A public domain dataset for human activity recognition using smartphones
A survey of old and new results for the test error estimation of a classifier
ABSTRACT The estimation of the generalization error of a trained classifier by means of a test se... more ABSTRACT The estimation of the generalization error of a trained classifier by means of a test set is one of the oldest problems in pattern recognition and machine learning. Despite this problem has been addressed for several decades, it seems that the last word has not been written yet, because new proposals continue to appear in the literature. Our objective is to survey and compare old and new techniques, in terms of quality of the estimation, easiness of use, and rigorousness of the approach, so to understand if the new proposals represent an effective improvement on old ones.
Out-of-Sample Error Estimation: The Blessing of High Dimensionality
2014 IEEE International Conference on Data Mining Workshop, 2014
Smartphone battery saving by bit-based hypothesis spaces and local Rademacher Complexities
2014 International Joint Conference on Neural Networks (IJCNN), 2014
Fully Empirical and Data-Dependent Stability-Based Bounds
IEEE Transactions on Cybernetics, 2014
The purpose of this paper is to obtain a fully empirical stability-based bound on the generalizat... more The purpose of this paper is to obtain a fully empirical stability-based bound on the generalization ability of a learning procedure, thus, circumventing some limitations of the structural risk minimization framework. We show that assuming a desirable property of a learning algorithm is sufficient to make data-dependency explicit for stability, which, instead, is usually bounded only in an algorithmic-dependent way. In addition, we prove that a well-known and widespread classifier, like the support vector machine (SVM), satisfies this condition. The obtained bound is then exploited for model selection purposes in SVM classification and tested on a series of real-world benchmarking datasets demonstrating, in practice, the effectiveness of our approach.
Machine Learning algorithms allow to create highly adaptable systems, since their functionality o... more Machine Learning algorithms allow to create highly adaptable systems, since their functionality only depends on the features of the inputs and the coefficients found during the training stage. In this paper, we present a method for building Support Vector Machines (SVM), characterized by integer parameters and coefficients. This method is useful to implement a pattern recognition system on resource-limited hardware, where a floatingpoint unit is often unavailable.
In the last years with the flourishing of the WSN (wireless sensor network) paradigm, ignited by ... more In the last years with the flourishing of the WSN (wireless sensor network) paradigm, ignited by DARPA funded UC Berkeley "Smart Dust" project, the monitoring and exploration of the terrestrial enviroment has greatly improved. The acquatic world, which covers more than the 70% of the earth, instead, has been largely unaffected by the WSN revolution due to the difficulty of transferring most of the knowhow developed for terrestrial and aerial systems and devices to their underwater counterparts. The aim of this article is to propose a new generation of UWSN(Underwater Wireless Sensor Network), called Smart Plankton, by getting inspiration from marine biology and acquatic microorganisms. ware, namely a microphone and a speaker and elastic latex membranes will be used to waterproof each device. New algorithm can be developed such as the hardware-friendly SVM described in [4] addressed problem of equalizing a nonlinear communication channel. Figure 3. Light absorption in water (picture from M.Chaplin, Water Structure and Science)
A common belief is that Machine Learning Theory (MLT) is not very useful, in pratice, for perform... more A common belief is that Machine Learning Theory (MLT) is not very useful, in pratice, for performing effective SVM model selection. This fact is supported by experience, because well-known hold-out methods like cross-validation, leave-one-out, and the bootstrap usually achieve better results than the ones derived from MLT. We show in this paper that, in a small sample setting, i.e. when the dimensionality of the data is larger than the number of samples, a careful application of the MLT can outperform other methods in selecting the optimal hyperparameters of a SVM.
In this paper, we focus the attention on one of the oldest problems in pattern recognition and ma... more In this paper, we focus the attention on one of the oldest problems in pattern recognition and machine learning: the estimation of the generalization error of a classifier through a test set. Despite this problem has been addressed for several decades, the last word has not yet been written, as new proposals continue to appear in the literature. Our objective is to survey and compare old and new techniques, in terms of quality of the estimation, easiness of use, and rigorousness of the approach.
Uploads
Papers by Alessandro Ghio