Conference Papers

Explainable Noisy Label Flipping for Multi-Label Fashion Image Classification

Ferreira B.; Costeira J., Gomes J.

CPVR21

2021

Abstract:

In online shopping applications, the daily insertion of new products requires an overwhelming annotation effort. Usually done by humans, it comes at a huge cost and yet generates high rates of noisy/missing labels that seriously hinder the effectiveness of CNNs in multi-label classification. We propose SELF-ML, a classification framework that exploits the relation between visual attributes and appearance together with the “low-rank” nature of the feature space. It learns a sparse reconstruction of image features as a convex combination of very few images – a basis – that are correctly annotated. Building on this representation, SELF-ML has a module that relabels noisy annotations from the derived combination of the clean data. Due to such structured reconstruction, SELF-ML gives an explanation of its label-flipping decisions. Experiments on a real-world shopping dataset show that SELF-ML significantly increases the number of correct labels even with few clean annotations.

Exploiting the height of vehicles in vehicular communication

Boban M., Meireles R., Barros J., Tonguz O., Steenkiste P.

IEEE Vehicular Networking Conference, VNC

2011

Abstract:

One of the most challenging research issues in vehicular ad hoc networks (VANETs) is how to efficiently relay messages between vehicles. We propose a heuristic that uses the physical dimensions of vehicles to help determine whether or not a vehicle is an appropriate next hop. We base the heuristic on the intuition that taller vehicles have an advantage over shorter ones because the former are less susceptible to shadowing from other vehicles. We implement a model that evaluates the efficacy of the proposed heuristic and we perform the experiments to validate the model. Based on both the experimental measurements and the simulations performed using the model, it is shown that tall vehicles consistently and significantly increase both the effective communication range and the message reachability. The effective communication range increased by more than 50%: from 290 meters when short vehicles are communicating to 450 meters in the case of tall vehicles. The results suggest that, when available, tall vehicles are significantly more likely to better relays than short vehicles. The proposed heuristic is not dependent on any specific routing technique and can be used to improve the performance of different classes of routing protocols.

Exploring events and distributed representations of text in multi-document summarization

Marujo L., Ling W., Ribeiro R., Gershman A., Carbonell J., Martins De Matos D., Neto J.P.

Knowledge-Based Systems

2016

Abstract:

In this article, we explore an event detection framework to improve multi-document summarization. Our approach is based on a two-stage single-document method that extracts a collection of key phrases, which are then used in a centrality-as-relevance passage retrieval model. We explore how to adapt this single-document method for multi-document summarization methods that are able to use event information. The event detection method is based on Fuzzy Fingerprint, which is a supervised method trained on documents with annotated event tags. To cope with the possible usage of different terms to describe the same event, we explore distributed representations of text in the form of word embeddings, which contributed to improve the summarization results. The proposed summarization methods are based on the hierarchical combination of single-document summaries. The automatic evaluation and human study performed show that these methods improve upon current state-of-the-art multi-document summarization systems on two mainstream evaluation datasets, DUC 2007 and TAC 2009. We show a relative improvement in ROUGE-1 scores of 16% for TAC 2009 and of 17% for DUC 2007.

Expressive lights for revealing mobile service robot state

Baraka K., Paiva A., Veloso M.

Advances in Intelligent Systems and Computing

2016

Abstract:

Autonomous mobile service robots move in our buildings, carrying out different tasks across multiple floors. While moving and performing their tasks, these robots find themselves in a variety of states. Although speech is often used for communicating the robot’s state to humans, such communication can often be ineffective. We investigate the use of lights as a persistent visualization of the robot’s state in relation to both tasks and environmental factors. Programmable lights offer a large degree of choices in terms of animation pattern, color and speed. We present this space of choices and introduce different animation profiles that we consider to animate a set of programmable lights on the robot. We conduct experiments to query about suitable animations for three representative scenarios of our autonomous symbiotic robot, CoBot. Our work enables CoBot to make its state persistently visible to humans.

Extending the metric dimension to graphs with missing edges

Zejnilovic S., Mitsche D., Gomes J., Sinopoli B.

Theoretical Computer Science

2016

Abstract:

The metric dimension of a connected graph G is the minimum number of vertices in a subset S of the vertex set of G such that all other vertices are uniquely determined by their distances to the vertices in S. We define an extended metric dimension for graphs with some edges missing, which corresponds to the minimum number of vertices in a subset S such that all other vertices have unique distances to S in all minimally connected graphs that result from completing the original graph. This extension allows for incomplete knowledge of the underlying graph in applications such as localizing the source of infection. We give precise values for the extended metric dimension when the original graph’s disconnected components are trees, cycles, grids, complete graphs, and we provide general upper bounds on this number in terms of the boundary of the graph.

Extracting latent structure from multiple interacting neural populations

Semedo J.D., Zandvakili A., Kohn A., Machens C.K., Yu B.M.

Advances in Neural Information Processing Systems

2014

Abstract:

Developments in neural recording technology are rapidly enabling the recording of populations of neurons in multiple brain areas simultaneously, as well as the identification of the types of neurons being recorded (e.g., excitatory vs. inhibitory). There is a growing need for statistical methods to study the interaction among multiple, labeled populations of neurons. Rather than attempting to identify direct interactions between neurons (where the number of interactions grows with the number of neurons squared), we propose to extract a smaller number of latent variables from each population and study how these latent variables interact. Specifically, we propose extensions to probabilistic canonical correlation analysis (pCCA) to capture the temporal structure of the latent variables, as well as to distinguish within-population dynamics from between-population interactions (termed Group Latent Auto-Regressive Analysis, gLARA). We then applied these methods to populations of neurons recorded simultaneously in visual areas V1 and V2, and found that gLARA provides a better description of the recordings than pCCA. This work provides a foundation for studying how multiple populations of neurons interact and how this interaction supports brain function.

Fast cooperative distributed learning

Jakovetic D., Moura J.M.F., Xavier J.

Conference Record - Asilomar Conference on Signals, Systems and Computers

2012

Abstract:

We consider distributed optimization where N agents in a network minimize the sum equation of their individual convex costs. To solve the described problem, existing literature proposes distributed gradient-like algorithms that are attractive due to computationally simple iterations k, but have a drawback of slow convergence (in k) to a solution. We propose a distributed gradient-like algorithm, that we build from the (centralized) Nesterov gradient method. For the convex f i ‘s with Lipschitz continuous and bounded gradients, we show that our method converges at rate O(log k/k). The achieved rate significantly improves over the convergence rate of existing distributed gradient-like methods, while the proposed algorithm maintains the same communication cost per k and a very similar computational cost per k. We further show that the rate O(log k/k) still holds if the bounded gradients assumption is replaced by a certain linear growth assumption. We illustrate the gains obtained by our method on two simulation examples: acoustic source localization and learning a linear classifier based on l 2 -regularized logistic loss.

Fast distributed gradient methods

Jakovetic D., Xavier J., Moura J.M.F.

IEEE Transactions on Automatic Control

2014

Abstract:

We study distributed optimization problems when N nodes minimize the sum of their individual costs subject to a common vector variable. The costs are convex, have Lipschitz continuous gradient (with constant L), and bounded gradient. We propose two fast distributed gradient algorithms based on the centralized Nesterov gradient algorithm and establish their convergence rates in terms of the per-node communications K and the per-node gradient evaluations k. Our first method, Distributed Nesterov Gradient, achieves rates O( logK/K) and O(logk/k). Our second method, Distributed Nesterov gradient with Consensus iterations, assumes at all nodes knowledge of L and μ(W) – the second largest singular value of the N ×N doubly stochastic weight matrix W. It achieves rates O( 1/ K 2-ξ ) and O( 1/k 2 ) ( ξ > 0 arbitrarily small). Further, we give for both methods explicit dependence of the convergence constants on N and W. Simulation examples illustrate our findings.

Fast incremental method for matrix completion: An application to trajectory correction

Cabral R.S., Costeira J.P., De La Torre F., Bernardino A.

Proceedings - International Conference on Image Processing, ICIP

2011

Abstract:

We address the problem of incrementally recovering a matrix of tracked image points, based on partial observations of their trajectories. Besides partial observability, we assume the existence of gross, but sparse, noise on the known entries. This problem has obvious applications in real-time tracking and structure from motion, where observations are plagued by self-occlusion and outliers. Recently, work in the optimization community has spun optimal methods for matrix completion when this matrix is known to be low rank by minimizing the nuclear norm, the sum of its singular values. Despite exhibiting several optimality properties, no available algorithms perform this minimization incrementally. In this paper, we build upon the Nuclear Norm Robust PCA method and SPectrally Optimal Completion to propose a fast and incremental algorithm which is able to cope with outliers. We present experiments showing the competitive speed of our method while maintaining performance comparable to the state-of-the-art.