One of the main problems in controlling COVID-19 epidemic spread is the delay in confirming cases. Having information on changes in the epidemic evolution or outbreaks rise before lab-confirmation is crucial in decision making for Public Health policies. We present an algorithm to estimate on-stream the number of COVID-19 cases using the data from telephone calls to a COVID-line. By modeling the calls as background (proportional to population) plus signal (proportional to infected), we fit the calls in Province of Buenos Aires (Argentina) with coefficient of determination $R^2 > 0.85$. This result allows us to estimate the number of cases given the number of calls from a specific district, days before the lab results are available. We validate the algorithm with real data. We show how to use the algorithm to track on-stream the epidemic, and present the Early Outbreak Alarm to detect outbreaks in advance to lab results. One key point in the developed algorithm is a detailed track of the uncertainties in the estimations, since the alarm uses the significance of the observables as a main indicator to detect an anomaly. We present the details of the explicit example in Villa Azul (Quilmes) where this tool resulted crucial to control an outbreak on time. The presented tools have been designed in urgency with the available data at the time of the development, and therefore have their limitations which we describe and discuss. We consider possible improvements on the tools, many of which are currently under development.
Few-shot learning aims to train models on a limited number of labeled samples given in a support set in order to generalize to unseen samples from a query set. In the standard setup, the support set contains an equal amount of data points for each class. However, this assumption overlooks many practical considerations arising from the dynamic nature of the real world, such as class-imbalance. In this paper, we present a detailed study of few-shot class-imbalance along three axes: meta-dataset vs. task imbalance, effect of different imbalance distributions (linear, step, random), and effect of rebalancing techniques. We extensively compare over 10 state-of-the-art few-shot learning and meta-learning methods using unbalanced tasks and meta-datasets. Our analysis using Mini-ImageNet reveals that 1) compared to the balanced task, the performances on class-imbalance tasks counterparts always drop, by up to $18.0\%$ for optimization-based methods, and up to $8.4$ for metric-based methods, 2) contrary to popular belief, meta-learning algorithms, such as MAML, do not automatically learn to balance by being exposed to imbalanced tasks during (meta-)training time, 3) strategies used to mitigate imbalance in supervised learning, such as oversampling, can offer a stronger solution to the class imbalance problem, 4) the effect of imbalance at the meta-dataset level is less significant than the effect at the task level with similar imbalance magnitude. The code to reproduce the experiments is released under an open-source license.
This paper presents a new deep learning-based model for fault detection in offshore wind turbines. To design a generalizable model for fault detection, we use 5 sensors and a sliding window to exploit the inherent temporal information contained in the raw time-series data obtained from sensors. The proposed model uses the nonlinear relationships among multiple sensor variables and the temporal dependency of each sensor on others that considerably increases the performance of fault detection model. A 10-fold cross-validation is used to verify the generalization of the model and evaluate the classification metrics. To evaluate the performance of the model, simulated data from a benchmark floating offshore wind turbine (FOWT) with supervisory control and data acquisition (SCADA) are used. The results illustrate that the proposed model would accurately disclose and classify more than 99% of the faults. Moreover, it is generalizable and can be used to detect faults for different types of systems.
Data-driven methods open up unprecedented possibilities for maritime surveillance using Automatic Identification System (AIS) data. In this work, we explore deep learning strategies using historical AIS observations to address the problem of predicting future vessel trajectories with a prediction horizon of several hours. We propose novel sequence-to-sequence vessel trajectory prediction models based on encoder-decoder recurrent neural networks (RNNs) that are trained on historical trajectory data to predict future trajectory samples given previous observations. The proposed architecture combines Long Short-Term Memory (LSTM) RNNs for sequence modeling to encode the observed data and generate future predictions with different intermediate aggregation layers to capture space-time dependencies in sequential data. Experimental results on vessel trajectories from an AIS dataset made freely available by the Danish Maritime Authority show the effectiveness of deep-learning methods for trajectory prediction based on sequence-to-sequence neural networks, which achieve better performance than baseline approaches based on linear regression or feed-forward networks. The comparative evaluation of results shows: i) the superiority of attention pooling over static pooling for the specific application, and ii) the remarkable performance improvement that can be obtained with labeled trajectories, i.e. when predictions are conditioned on a low-level context representation encoded from the sequence of past observations, as well as on additional inputs (e.g., the port of departure or arrival) about the vessel's high-level intention which may be available from AIS.
We study the labeled multi-robot path planning problem in continuous 2D and 3D domains in the absence of obstacles where robots must not collide with each other. For an arbitrary number of robots in arbitrary initial and goal arrangements, we derive a polynomial time, complete algorithm that produces solutions with constant-factor optimality guarantees on both makespan and distance optimality, in expectation, under the assumption that the robot labels are uniformly randomly distributed. Our algorithm only requires a small constant factor expansion of the initial and goal configuration footprints for solving the problem, i.e., the problem can be solved in a fairly small bounded region. Beside theoretical guarantees, we present a thorough computational evaluation of the proposed solution. In addition to the baseline implementation, adapting an effective (but non-polynomial time) routing subroutine, we also provide a highly efficient implementation that quickly computes near-optimal solutions. Hardware experiments on the microMVP platform composed of non-holonomic robots confirms the practical applicability of our algorithmic pipeline.
We propose advances that address two key challenges in future trajectory prediction: (i) multimodality in both training data and predictions and (ii) constant time inference regardless of number of agents. Existing trajectory predictions are fundamentally limited by lack of diversity in training data, which is difficult to acquire with sufficient coverage of possible modes. Our first contribution is an automatic method to simulate diverse trajectories in the top-view. It uses pre-existing datasets and maps as initialization, mines existing trajectories to represent realistic driving behaviors and uses a multi-agent vehicle dynamics simulator to generate diverse new trajectories that cover various modes and are consistent with scene layout constraints. Our second contribution is a novel method that generates diverse predictions while accounting for scene semantics and multi-agent interactions, with constant-time inference independent of the number of agents. We propose a convLSTM with novel state pooling operations and losses to predict scene-consistent states of multiple agents in a single forward pass, along with a CVAE for diversity. We validate our proposed multi-agent trajectory prediction approach by training and testing on the proposed simulated dataset and existing real datasets of traffic scenes. In both cases, our approach outperforms SOTA methods by a large margin, highlighting the benefits of both our diverse dataset simulation and constant-time diverse trajectory prediction methods.
Convolutional neural networks (CNN) exhibit unmatched performance in a multitude of computer vision tasks. However, the advantage of using convolutional networks over fully-connected networks is not understood from a theoretical perspective. In this work, we show how convolutional networks can leverage locality in the data, and thus achieve a computational advantage over fully-connected networks. Specifically, we show a class of problems that can be efficiently solved using convolutional networks trained with gradient-descent, but at the same time is hard to learn using a polynomial-size fully-connected network.
Automatic text summarization has enjoyed great progress over the last years. Now is the time to re-assess its focus and objectives. Does the current focus fully adhere to users' desires or should we expand or change our focus? We investigate this question empirically by conducting a survey amongst heavy users of pre-made summaries. We find that the current focus of the field does not fully align with participants' wishes. In response, we identify three groups of implications. First, we argue that it is important to adopt a broader perspective on automatic summarization. Based on our findings, we illustrate how we can expand our view when it comes to the types of input material that is to be summarized, the purpose of the summaries and their potential formats. Second, we define requirements for datasets that can facilitate these research directions. Third, usefulness is an important aspect of summarization that should be included in our evaluation methodology; we propose a methodology to evaluate the usefulness of a summary. With this work we unlock important research directions for future work on automatic summarization and we hope to initiate the development of methods in these directions.
Temporal relational modeling in video is essential for human action understanding, such as action recognition and action segmentation. Although Graph Convolution Networks (GCNs) have shown promising advantages in relation reasoning on many tasks, it is still a challenge to apply graph convolution networks on long video sequences effectively. The main reason is that large number of nodes (i.e., video frames) makes GCNs hard to capture and model temporal relations in videos. To tackle this problem, in this paper, we introduce an effective GCN module, Dilated Temporal Graph Reasoning Module (DTGRM), designed to model temporal relations and dependencies between video frames at various time spans. In particular, we capture and model temporal relations via constructing multi-level dilated temporal graphs where the nodes represent frames from different moments in video. Moreover, to enhance temporal reasoning ability of the proposed model, an auxiliary self-supervised task is proposed to encourage the dilated temporal graph reasoning module to find and correct wrong temporal relations in videos. Our DTGRM model outperforms state-of-the-art action segmentation models on three challenging datasets: 50Salads, Georgia Tech Egocentric Activities (GTEA), and the Breakfast dataset. The code is available at https://github.com/redwang/DTGRM.
Recognition in planning seeks to find agent intentions, goals or activities given a set of observations and a knowledge library (e.g. goal states, plans or domain theories). In this work we introduce the problem of Online Action Recognition. It consists in recognizing, in an open world, the planning action that best explains a partially observable state transition from a knowledge library of first-order STRIPS actions, which is initially empty. We frame this as an optimization problem, and propose two algorithms to address it: Action Unification (AU) and Online Action Recognition through Unification (OARU). The former builds on logic unification and generalizes two input actions using weighted partial MaxSAT. The latter looks for an action within the library that explains an observed transition. If there is such action, it generalizes it making use of AU, building in this way an AU hierarchy. Otherwise, OARU inserts a Trivial Grounded Action (TGA) in the library that explains just that transition. We report results on benchmarks from the International Planning Competition and PDDLGym, where OARU recognizes actions accurately with respect to expert knowledge, and shows real-time performance.