Conference Papers

Supervised Hyperspectral Image Classification with Rejection

Condessa F., Bioucas-Dias J., Kovacevic J.

IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing

2016

Abstract:

Hyperspectral image classification is a challenging problem as obtaining complete and representative training sets is costly, pixels can belong to unknown classes, and it is generally an ill-posed problem. The need to achieve high classification accuracy may surpass the need to classify the entire image. To account for this scenario, we use classification with rejection by providing the classifier with an option not to classify a pixel and consequently reject it. We present and analyze two approaches for supervised hyperspectral image classification that combine the use of contextual priors with classification with rejection: 1) by jointly computing context and rejection and 2) by sequentially computing context and rejection. In the joint approach, rejection is introduced as an extra class that models the probability of classifier failure. In the sequential approach, rejection results from the hidden field associated with a marginal maximum a posteriori classification of the image. We validate both approaches on real hyperspectral data.

Teaching the Machine to Explain Itself using Domain Knowledge

Balayan V., Saleiro P., Belém C., Krippahl L., Bizarro P.

NeurIPS’2020/HAMLETS

2020

Abstract:

Machine Learning (ML) has been increasingly used to aid humans to make better and faster decisions. However, non-technical humans-in-the-loop struggle to comprehend the rationale behind model predictions, hindering trust in algorithmic decision-making systems. Considerable research work on AI explainability attempts to win back trust in AI systems by developing explanation methods but there is still no major breakthrough. At the same time, popular explanation methods (e.g., LIME, and SHAP) produce explanations that are very hard to understand for non-data scientist persona. To address this, we present JOEL, a neural network-based framework to jointly learn a decision-making task and associated explanations that convey domain knowledge. JOEL is tailored to human-in-the-loop domain experts that lack deep technical ML knowledge, providing high-level insights about the model’s predictions that very much resemble the experts’ own reasoning. Moreover, we collect the domain feedback from a pool of certified experts and use it to ameliorate the model (human teaching), hence promoting seamless and better suited explanations. Lastly, we resort to semantic mappings between legacy expert systems and domain taxonomies to automatically annotate a bootstrap training set, overcoming the absence of concept-based human annotations. We validate JOEL empirically on a real-world fraud detection dataset. We show that JOEL can generalize the explanations from the bootstrap dataset. Furthermore, obtained results indicate that human teaching can further improve the explanations prediction quality by approximately 13.57%.

Technology Forgiveness: Why emerging technologies differ in their resilience to institutional instability

Roca J.; Vaishnav P.; Morgan G.; Fuchs E.; Mendonça J.

Technological Forecasting and Social Change

2021

Abstract:

Long-term public support may encourage the diffusion of emerging technologies by coordinating the generation of knowledge and providing patient funding, but unexpected policy changes may hinder private investment and even lead to situations of technology lockout. Leveraging archival data; insights from 45 interviews across academia, industry, and government; and 75 hours of participant observations, we develop insights about why institutional instability in Portugal affected the adoption of Polymer Additive Manufacturing (PAM) and Metal Additive Manufacturing (MAM) differently. In both cases, Portugal invested in the technology relatively early. While PAM has been widely adopted, including increasingly in high-tech applications, MAM adoption has been modest despite MAM’s potential to greatly improve the performance and competitiveness of metal molds. From the comparison between PAM and MAM, we generate theory about technological and contextual factors that affect ‘technological forgiveness’, defined as the resilience of a new technology’s adoption to institutional instability.

Terahertz Massive MIMO for Beyond-5G Wireless Communication

Busari S., Huq K., Mumtaz S., Rodriguez J.

ICC 2019 - 2019 IEEE International Conference on Communications (ICC)

2019

Abstract:

Spectrum use will undoubtedly move to the terahertz (THz) frequencies in the beyond fifth-generation (B5G) mobile system era. With enormous bandwidth far greater than the amount available in the microwave and millimeter-wave bands combined, THz communication will open up new frontiers for exciting services and applications requiring ultra-broadband connectivity. In this work, we evaluate the performance of a candidate B5G scenario with THz-enabled massive MIMO access points mounted on street lampposts to serve pedestrian users. Using spectral efficiency (SE) and energy efficiency (EE) as metrics, we compared the performance of three precoding schemes, namely: analog-only beamsteering, hybrid precoding with baseband zero forcing and singular value decomposition precoding as upper bound. We also show the impacts of carrier frequency, bandwidth and antenna gain on the system performance. The simulation results reveal the optimal EE and SE points which are critical design goals for the green and sustainable operation of next-generation networks.

Text-dependent pathological voice detection

Anumanchipalli G.K., Meinedo H., Bugalho M., Trancoso I., Oliveira L.C., Black A.W.

13th Annual Conference of the International Speech Communication Association 2012, INTERSPEECH 2012

2012

Abstract:

While global characteristics of the speaker’s source and spectral features have been successfully employed in pathological voice detection, the underlying text has largely been ignored. In this work, we focus on experiments that exploit the text stimulus that is read by the subject. Features derived from text include the mean cepstral distortion of the subject from an average intelligible speaker, and prosodic features include the speaking rate, statistics of phoneme durations, etc. The phonetic labeling information is also exploited to ignore all the unvoiced regions of the speech samples to improve the discriminability between intelligible and pathological voices. We also designed features that capture the speaker’s overall closeness to intelligible instances of the same text stimulus from other speakers. Our experiments show that the proposed text-derived features improve the detection of pathological voices by 20%. Index Terms: Pathological voices, example based detection, text-driven features, fusion of classification methods.

Textual event detection using fuzzy fingerprints

Marujo L., Carvalho J.P., Gershman A., Carbonell J., Neto J.P., Martins de Matos D.

Advances in Intelligent Systems and Computing

2015

Abstract:

In this paper we present a method to improve the automatic detection of events in short sentences when in the presence of a large number of event classes. Contrary to standard classification techniques such as Support Vector Machines or Random Forest, the proposed Fuzzy Fingerprints method is able to detect all the event classes present in the ACE 2005 Multilingual Corpus, and largely improves the obtained G-Mean value.

The cloud inside the network: A virtualization approach to resource allocation

Soares J., Monteiro R., Melo M., Sargento S., Carapinha J.

Cloud Technology: Concepts, Methodologies, Tools, and Applications

2014

Abstract:

The access infrastructure to the cloud is usually a major drawback that limits the uptake of cloud services. Attention has turned to rethinking a new architectural deployment of the overall cloud service delivery. In this chapter, the authors argue that it is not sufficient to integrate the cloud domain with the operator’s network domain based on the current models. They envision a full integration of cloud and network, where cloud resources are no longer confined to a data center but are spread throughout the network and owned by the network operator. In such an environment, challenges arise at different levels, such as in resource management, where both cloud and network resources need to be managed in an integrated approach. The authors particularly address the resource allocation problem through joint virtualization of network and cloud resources by studying and comparing an Integer Linear Programming formulation and a heuristic algorithm.

The Economics of Multi-Network Access

Zhang N, Sirbu A.M., Peha J.M.

2019 IEEE International Symposium on Dynamic Spectrum Access Networks (DySPAN)

2019

Abstract:

Traditionally, a cell phone remains on a single, primary mobile network operator (MNO) as long as it is available, and roams onto another MNO only when outside the primary MNO’s coverage. Multi-network access (MNA) is a new scheme where a cell phone may use any one of multiple MNOs at any place, anytime. One such example is a multi-operator mobile virtual network operator (MO-MVNO) like Google Fi. This paper quantifies how much MNA can reduce the cost of cellular data services, and shows that the amount of infrastructure and/or spectrum resources needed to produce a given network capacity can be reduced by over 20%. Greater resource savings can be realized if MNA-capable devices attach to towers of higher SINR rather than higher expected data rate. The amount of resources saved increases faster than linearly with increasing fraction of MNA-capable devices on the network, so as an MO-MVNO gains market share, it could demand better wholesale prices from partner MNOs. If the distribution of traffic volume between partner MNOs shifts significantly with MNA, an MNO losing traffic share may not have an incentive to participate in MNA unless it could demand a much higher wholesale price than other partner MNOs, possibly close to or even above the retail price net of market cost. The eventual economic impacts on each operator adopting MNA are the result of complex considerations involving not only business decisions like investment and wholesale pricing, but also technical parameters like network selection algorithms and resource allocation schemes.

The effectiveness of decarbonizing the passenger transport sector through monetary incentives

Santarromana R.; Mendonça J.; Dias A.M.

Transportation Research Part A: Policy and Practice

2016

Abstract:

Passenger cars account for most road transportation emissions, and almost half of overall transport sector emissions in the EU. Countries in Europe have established policies to achieve emissions reductions in the transport sector by incentivizing the acquisition of fuel-efficient vehicles. In this paper, we perform a pair-wise comparison of common passenger vehicles sold in 2017, which implements newer data and more realistic assumptions than an earlier study. The pair-wise study compares an electric vehicle (EV) against a similar combustion vehicle to simulate a real market choice for consumers—a method commonly used to elicit preferences—and shows that fiscal incentives are effective at increasing EV acquisition. Acquiring EVs over conventional vehicles alone contributes to about a 60% reduction per kilometer of well-to-wheel emissions, based on average emissions of new EU vehicle fleets in 2017. A second mechanism at reducing emissions in the transport sector is through incentivizing consumer charging behavior to use less carbon intense electricity. The electricity used to charge EVs is variable throughout a day; therefore, we propose a dynamic pricing mechanism dependent on the carbon intensity of the electricity grid. We do this analysis through a case study for Portugal using the entire country’s public charging demands from 2017. The responsiveness of the users to the variable price is reflected by the market price elasticity of demand, and the resulting reduction in demand from the surcharge is approximated. Our study finds that a surcharging mechanism based on the carbon intensity of the electric grid can yield an emissions reduction of 20 tonnes per year while still achieving profits.