Teaching iCub to recognize objects using deep Convolutional Neural Networks

By Giulia Pasquale, Carlo Ciliberto, Francesca Odone, Lorenzo Rosasco, Lorenzo Natale. 2015

55 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

Is Deep Learning Safe for Robot Vision? Adversarial Examples Against the iCub Humanoid (2017, methodology): “The linear classifiers used for this purpose include Support Vector Machines (SVMs) and Recursive Least Square (RLS) classifiers, as both can be efficiently updated online [21].”
CORe50: a New Dataset and Benchmark for Continuous Object Recognition (2017, methodology): “Finally, the iCubWorld datasets ([19], [20]) have been acquired in a robotic vision context and are the closest ones to CORe50 .”
Continual Learning for Robotics (2019, methodology): “Initial progresses in this area have been proposed in (Thrun and Mitchell, 1995; Pasquale et al., 2015; Lomonaco and Maltoni, 2016; Camoriano et al., 2017; Lomonaco and Maltoni, 2017).”
Rehearsal-Free Continual Learning over Small Non-I.I.D. Batches (2019, background): “Objects are presented to the robot by a human operator who can also provide the labels, thus enabling a supervised classification (such an applicative scenario is well described in [28, 26, 27]).”
Incremental On-Line Learning of Object Classes using a Combination of Self-Organizing Incremental Neural Networks and Deep Convolutional Neural Networks* (2016, methodology): “Pasquale et al. [9] investigated how to leverage the success of deep learning architectures in an incremental setting by using the extracted features from a pre-trained Convolutional Neural Network (CNN) as input to a linear classiﬁer that can be updated every time new samples become available.”
Video Object Segmentation using Teacher-Student Adaptation in a Human Robot Interaction (HRI) Setting (2018, unknown): “RGB-D [25] - 51 300 T BIG BIRD [26] - - 100 T ICUB 28 [20] 4 7 28 H ICUB World [21] 6 20 200 H Core50 [17] 11 10 50 H IVOS 12 12 36 H”
Lifelong Learning and Personalization in Long-Term Human-Robot Interaction (LEAP-HRI) (2021, background): “Incremental learning and adaptation based on continuous stream of observations are needed in various settings in robotics, such as object, user or facial expression recognition [1], [7]–[13], motion”
Incremental robot learning of new objects with fixed update time (2016, methodology): “We performed feature extraction as specified in [30], i.”
Incremental online learning of objects for robots operating in real environments (2017, background): “[15] addressed the problem of incremental learning but from a different perspective.”
Incremental Object Recognition in Robotics with Extension to New Classes in Constant Time (2016, background): “This would lead to O(dkT ) computations at each iteration, which is undesirable due to the dependency on k.”
Online Continual Learning on Sequences (2020, methodology): “Finally, iCubWorld ([65], [66]) and OpenLORIS have been acquired in a robotic vision context and are similar to CORe50.”
Object identification from few examples by improving the invariance of a Deep Convolutional Neural Network (2016, background): “Their learning rates (namely the step-size used by the stochastic gradient descent during back-propagation) can therefore be set to zero (no update is done) or to very small values, while the learning rate of the output layer is maintained to a higher value.”
Fine-Grained Continual Learning (2019, background): “…to leave some room to accommodate all their training sessions. for each run in num _ runs : while not assigned : assigned = true robot by a human operator who can also provide the labels, thus enabling a supervised classiﬁcation (such an applicative scenario is well described in [26, 27, 28]).”