Imperial College London

DrCarloCiliberto

Faculty of EngineeringDepartment of Electrical and Electronic Engineering

Lecturer
 
 
 
//

Contact

 

+44 (0)20 7594 6173c.ciliberto CV

 
 
//

Location

 

1003Electrical EngineeringSouth Kensington Campus

//

Summary

 

Publications

Publication Type
Year
to

38 results found

Fanello SR, Ciliberto C, Santoro M, Natale L, Metta G, Rosasco L, Odone Fet al., 2013, iCub World: Friendly Robots Help Building Good Vision Data-Sets, 2013 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION WORKSHOPS (CVPRW), Pages: 700-705, ISSN: 2160-7508

Journal article

Ciliberto C, Fanello SR, Natale L, Metta Get al., 2012, A Heteroscedastic Approach to Independent Motion Detection for Actuated Visual Sensors, 25th IEEE\RSJ International Conference on Intelligent Robots and Systems (IROS), Publisher: IEEE, Pages: 3907-3913, ISSN: 2153-0858

Conference paper

Ciliberto C, Smeraldi F, Natale L, Metta Get al., 2011, Online Multiple Instance Learning Applied to Hand Detection in a Humanoid Robot, IEEE/RSJ International Conference on Intelligent Robots and Systems, Publisher: IEEE, Pages: 1526-1532, ISSN: 2153-0858

Conference paper

Ciliberto C, Pattacini U, Natale L, Nori F, Metta Get al., 2011, Reexamining Lucas-Kanade Method for Real-Time Independent Motion Detection: Application to the iCub Humanoid Robot, IEEE/RSJ International Conference on Intelligent Robots and Systems, Publisher: IEEE, Pages: 4154-4160, ISSN: 2153-0858

Conference paper

Denevi G, Pontil M, Ciliberto C, The Advantage of Conditional Meta-Learning for Biased Regularization and Fine-Tuning

Biased regularization and fine-tuning are two recent meta-learningapproaches. They have been shown to be effective to tackle distributions oftasks, in which the tasks' target vectors are all close to a commonmeta-parameter vector. However, these methods may perform poorly onheterogeneous environments of tasks, where the complexity of the tasks'distribution cannot be captured by a single meta-parameter vector. We addressthis limitation by conditional meta-learning, inferring a conditioning functionmapping task's side information into a meta-parameter vector that isappropriate for that task at hand. We characterize properties of theenvironment under which the conditional approach brings a substantial advantageover standard meta-learning and we highlight examples of environments, such asthose with multiple clusters, satisfying these properties. We then propose aconvex meta-algorithm providing a comparable advantage also in practice.Numerical experiments confirm our theoretical findings.

Journal article

Marconi GM, Rosasco L, Ciliberto C, Hyperbolic Manifold Regression

Geometric representation learning has recently shown great promise in severalmachine learning settings, ranging from relational learning to languageprocessing and generative models. In this work, we consider the problem ofperforming manifold-valued regression onto an hyperbolic space as anintermediate component for a number of relevant machine learning applications.In particular, by formulating the problem of predicting nodes of a tree as amanifold regression task in the hyperbolic space, we propose a novelperspective on two challenging tasks: 1) hierarchical classification via labelembeddings and 2) taxonomy extension of hyperbolic representations. To addressthe regression problem we consider previous methods as well as proposing twonovel approaches that are computationally more advantageous: a parametric deeplearning model that is informed by the geodesics of the target space and anon-parametric kernel-method for which we also prove excess risk bounds. Ourexperiments show that the strategy of leveraging the hyperbolic geometry ispromising. In particular, in the taxonomy expansion setting, we find that thehyperbolic-based estimators significantly outperform methods performingregression in the ambient Euclidean space.

Journal article

Ciliberto C, Stamos D, Pontil M, Reexamining Low Rank Matrix Factorization for Trace Norm Regularization

Trace norm regularization is a widely used approach for learning low rankmatrices. A standard optimization strategy is based on formulating the problemas one of low rank matrix factorization which, however, leads to a non-convexproblem. In practice this approach works well, and it is often computationallyfaster than standard convex solvers such as proximal gradient methods.Nevertheless, it is not guaranteed to converge to a global optimum, and theoptimization can be trapped at poor stationary points. In this paper we showthat it is possible to characterize all critical points of the non-convexproblem. This allows us to provide an efficient criterion to determine whethera critical point is also a global minimizer. Our analysis suggests an iterativemeta-algorithm that dynamically expands the parameter space and allows theoptimization to escape any non-global critical point, thereby converging to aglobal minimizer. The algorithm can be applied to problems such as matrixcompletion or multitask learning, and our analysis holds for any randominitialization of the factor matrices. Finally, we confirm the good performanceof the algorithm on synthetic and real datasets.

Journal article

Pasquale G, Ciliberto C, Odone F, Rosasco L, Natale Let al., Real-world Object Recognition with Off-the-shelf Deep Conv Nets: How Many Objects can iCub Learn?

The ability to visually recognize objects is a fundamental skill for roboticssystems. Indeed, a large variety of tasks involving manipulation, navigation orinteraction with other agents, deeply depends on the accurate understanding ofthe visual scene. Yet, at the time being, robots are lacking good visualperceptual systems, which often become the main bottleneck preventing the useof autonomous agents for real-world applications. Lately in computer vision, systems that learn suitable visual representationsand based on multi-layer deep convolutional networks are showing remarkableperformance in tasks such as large-scale visual recognition and imageretrieval. To this regard, it is natural to ask whether such remarkableperformance would generalize also to the robotic setting. In this paper we investigate such possibility, while taking further steps indeveloping a computational vision system to be embedded on a robotic platform,the iCub humanoid robot. In particular, we release a new dataset ({\sciCubWorld28}) that we use as a benchmark to address the question: {\it how manyobjects can iCub recognize?} Our study is developed in a learning frameworkwhich reflects the typical visual experience of a humanoid robot like the iCub.Experiments shed interesting insights on the strength and weaknesses of currentcomputer vision approaches applied in real robotic settings.

Journal article

This data is extracted from the Web of Science and reproduced under a licence from Thomson Reuters. You may not copy or re-distribute this data in whole or in part without the written consent of the Science business of Thomson Reuters.

Request URL: http://wlsprd.imperial.ac.uk:80/respub/WEB-INF/jsp/search-html.jsp Request URI: /respub/WEB-INF/jsp/search-html.jsp Query String: id=01002937&limit=30&person=true&page=2&respub-action=search.html