Abstract

Abstract
Deep learning has offered significant performance improvements on several pattern recognition problems. However, the well-known need for large amounts of labeled data limits applicability and performance where those are not available. Hence, this paper proposes an adaptation of the triplet loss for self-learning with entirely unlabeled data, where there is uncertainty in the generated triplets. The methodology was applied to off-the-person electrocardiogram-based biometric authentication and unconstrained face identity verification tasks, including stress experiments designed to simulate more difficult circumstances. Despite the uncertainty related to the use of unlabeled data, the method was mostly capable of avoiding negatively affecting the model's performance. The promising results show the proposed method can be a viable alternative to supervised learning in cases where only unlabeled data are available. The method is especially suitable for training with continuous stream-based datasets such as on person re-identification in video streams and continuous electrocardiogram-based biometrics.

Abstract
A key to the generalization ability of Convolutional Neural Networks (CNNs) is the idea that patterns that appear in one region of the image have a high probability of appearing in other regions. This notion is also true for other spatial relationships, such as orientation. Motivated by the fact that in the early layers of CNNs distinct filters often encode for the same feature at different angles, we propose to incorporate the rotation equivariant prior in these models. In this work, different regularization strategies that capture the notion of approximate equivariance were designed and quantitatively evaluated in their ability to generate rotation-equivariant models and their effect on the model's capacity to generalize to unseen data. Some of these strategies consistently lead to higher test set accuracies when compared to a baseline model, on classification tasks. We conclude that the rotation equivariance prior should be adopted in the general setting when modeling visual data.

Abstract
The literature seems to indicate that the QRS complex is the most important component of the electrocardiogram (ECG) for biometrics. To verify this claim, we use interpretability tools to explain how a convolutional neural network uses ECG signals to identify people, using on-the-person (PTB) and off-the-person (UofTDB) signals. While the QRS complex appears indeed to be a key feature on ECG biometrics, especially with cleaner signals, results indicate that, for larger populations in off-the-person settings, the QRS shares relevance with other heartbeat components, which it is essential to locate. These insights indicate that avoiding excessive focus on the QRS complex, using decision explanations during training, could be useful for model regularisation.

Abstract
Importance Mammography screening currently relies on subjective human interpretation. Artificial intelligence (AI) advances could be used to increase mammography screening accuracy by reducing missed cancers and false positives. Objective To evaluate whether AI can overcome human mammography interpretation limitations with a rigorous, unbiased evaluation of machine learning algorithms. Design, Setting, and Participants In this diagnostic accuracy study conducted between September 2016 and November 2017, an international, crowdsourced challenge was hosted to foster AI algorithm development focused on interpreting screening mammography. More than 1100 participants comprising 126 teams from 44 countries participated. Analysis began November 18, 2016. Main Outcomes and Measurements Algorithms used images alone (challenge 1) or combined images, previous examinations (if available), and clinical and demographic risk factor data (challenge 2) and output a score that translated to cancer yes/no within 12 months. Algorithm accuracy for breast cancer detection was evaluated using area under the curve and algorithm specificity compared with radiologists' specificity with radiologists' sensitivity set at 85.9% (United States) and 83.9% (Sweden). An ensemble method aggregating top-performing AI algorithms and radiologists' recall assessment was developed and evaluated. Results Overall, 144 & x202f;231 screening mammograms from 85 & x202f;580 US women (952 cancer positive <= 12 months from screening) were used for algorithm training and validation. A second independent validation cohort included 166 & x202f;578 examinations from 68 & x202f;008 Swedish women (780 cancer positive). The top-performing algorithm achieved an area under the curve of 0.858 (United States) and 0.903 (Sweden) and 66.2% (United States) and 81.2% (Sweden) specificity at the radiologists' sensitivity, lower than community-practice radiologists' specificity of 90.5% (United States) and 98.5% (Sweden). Combining top-performing algorithms and US radiologist assessments resulted in a higher area under the curve of 0.942 and achieved a significantly improved specificity (92.0%) at the same sensitivity. Conclusions and Relevance While no single AI algorithm outperformed radiologists, an ensemble of AI algorithms combined with radiologist assessment in a single-reader screening environment improved overall accuracy. This study underscores the potential of using machine learning methods for enhancing mammography screening interpretation. Question How do deep learning algorithms perform compared with radiologists in screening mammography interpretation? Findings In this diagnostic accuracy study using 144 & x202f;231 screening mammograms from 85 & x202f;580 women from the United States and 166 & x202f;578 screening mammograms from 68 & x202f;008 women from Sweden, no single artificial intelligence algorithm outperformed US community radiologist benchmarks; including clinical data and prior mammograms did not improve artificial intelligence performance. However, combining best-performing artificial intelligence algorithms with single-radiologist assessment demonstrated increased specificity. Meaning Integrating artificial intelligence to mammography interpretation in single-radiologist settings could yield significant performance improvements, with the potential to reduce health care system expenditures and address resource scarcity experienced in population-based screening programs. This diagnostic accuracy study evaluates whether artificial intelligence can overcome human mammography interpretation limits with a rigorous, unbiased evaluation of machine learning algorithms.

Abstract
Video summarization is an important tool considering the amount of data to analyze. Techniques in this area aim to yield synthetic and useful visual abstraction of the videos contents. Hence, in this paper we present a new summarization algorithm, based on image features, which is composed by the following steps: (i) Query video processing using cosine similarity metric and total variation smoothing to identify classes in the query sequence; (ii) With this result, build a labeled training set of frames; (iii) Generate the unlabeled training set composed by samples of the video database; (iv) Training a deep semi-supervised autoencoder; (v) Compute the K-means for each video separately, in the encoder space, with the number of clusters set as a percentage of the video size; (vi) Select key-frames in the K-means clusters to define the summaries. In this methodology, the query video is used to incorporate prior knowledge in the whole process through the obtained labeled data. The step (iii) aims to include unknown patterns useful for the summarization process. We evaluate the methodology using some videos from OPV video database. We compare the performance of our algorithm with the VSum. The results indicate that the pipeline was well succeed in the summarization presenting a F-score value superior to VSum.