An assumption often made in supervised learning is that the training and testing sets have the same label distribution. However, in real-life scenarios, this assumption rarely holds. For example, medical diagnosis result distributions change over time and across locations; fraud detection models must adapt as patterns of fraudulent activity shift; the category distribution of social media posts changes based on trending topics and user demographics. In the task of label shift estimation, the goal is to estimate the changing label distribution pt(y) in the testing set, assuming the likelihood p(x|y) does not change, implying no concept drift. In this paper, we propose a new approach for post-hoc label shift estimation, unlike previous methods that perform moment matching with confusion matrix estimated from a validation set or maximize the likelihood of the new data with an expectation-maximization algorithm. We aim to incrementally update the prior on each sample, adjusting each posterior for more accurate label shift estimation. The proposed method is based on intuitive assumptions on classifiers that are generally true for modern probabilistic classifiers. The proposed method relies on a weaker notion of calibration compared to other methods. As a post-hoc approach for label shift estimation, the proposed method is versatile and can be applied to any black-box probabilistic classifier. Experiments on CIFAR-10 and MNIST show that the proposed method consistently outperforms the current state-of-the-art maximum likelihood-based methods under different calibrations and varying intensities of label shift.

Decentralized machine learning, such as Federated Learning (FL), is widely adopted in many application domains. Especially in domains like recommendation systems, sharing gradients instead of private data has recently caught the research community’s attention. Personalized travel route recommendation utilizes users’ location data to recommend optimal travel routes. Location data is extremely privacy sensitive, presenting increased risks of exposing behavioral patterns and demographic attributes. FL for route recommendation can mitigate the sharing of location data. However, this article shows that an adversary can recover the user trajectories used to train the federated recommendation models with high proximity accuracy. To this effect, we propose a novel attack called DeepSneak, which uses shared gradients obtained from global model training in FL to reconstruct private user trajectories. We formulate the attack as a regression problem and train a generative model by minimizing the distance between gradients. We validate the success of DeepSneak on two real-world trajectory datasets. The results show that we can recover the location trajectories of users with reasonable spatial and semantic accuracy.

The recent outbreak of coronavirus disease has demonstrated that physical human interactions and modern movement paradigms are the principle drivers for the rapid spatial spread of infectious diseases. Modelling the impact of human mobility is crucial to understand the underlying dynamics of disease spread and consequently to develop effective containment and control strategies. While previous studies have investigated the impact of specific mobility profiles on the spreading dynamics of infectious diseases, they used either highly aggregated spatio-temporal data or portions of datasets that span a short period of time. These limitations do not allow to study how the influence of different mobility aspects on the spread changes as a disease outbreak progresses. In this paper we use large-scale comprehensive human mobility traces to study the impact of the latent period on the spreading dynamics of diseases. In addition, we provide a detailed analysis of how the spreading power of different mobility profiles changes over time. We propose an approach that analyses the behaviour of the individuals' spreading power as time progresses. Through extensive disease spread simulations we uncover a population influence homogeneity threshold, defined by a percentage of the population at which the identified mobility groups become equally influential to the spread.

Human Activity Recognition (HAR) plays an irreplaceable role in various applications such as security, gaming, and assisted living. Recent studies introduce deep learning to mitigate the manual feature extraction (i.e., data representation) efforts and achieve high accuracy. However, there are still challenges in learning accurate representations for sensory data due to the weakness of representation modules and the subject variances. We propose a scheme called Distance-based HAR from Ensembled spatial-temporal Representations (DHARER) to address above challenges. The idea behind DHARER is straightforward—the same activities should have similar representations. We first learn representations of the input sensory segments and latent prototype representations of each class, using a Convolution Neural Network (CNN)-based dual-stream representation module; then the learned representations are projected to activity types by measuring their similarity to the learned prototypes. We have conducted extensive experiments under a strict subject-independent setting on three large-scale datasets to evaluate the proposed scheme, and our experimental results demonstrate superior performance of DHARER to several state-of-the-art methods.

Online ride-sharing platforms have become a critical part of the urban transportation system. Accurately recommending hotspots to drivers in such platforms is essential to help drivers find passengers and improve users' experience, which calls for efficient passenger demand prediction strategy. However, predicting multi-step passenger demand is challenging due to its high dynamicity, complex dependencies along spatial and temporal dimensions, and sensitivity to external factors (meteorological data and time meta). We propose an end-to-end deep learning framework to address the above problems. Our model comprises three components in pipeline: 1) a cascade graph convolutional recurrent neural network to accurately extract the spatial-temporal correlations within citywide historical passenger demand data; 2) two multi-layer LSTM networks to represent the external meteorological data and time meta, respectively; 3) an encoder-decoder module to fuse the above two parts and decode the representation to predict over multi-steps into the future. The experimental results on three real-world datasets demonstrate that our model can achieve accurate prediction and outperform the most discriminative state-of-the-art methods.

The recent wave of sensor-rich, Internet-enabled, smart mobile devices such as the Apple iPhone has opened the door for a novel paradigm for monitoring the urban landscape known as participatory sensing. Using this paradigm, ordinary citizens can collect multi-modal data streams from the surrounding environment using their mobile devices and share the same using existing communication infrastructure (e.g., 3G service or WiFi access points). The data contributed from multiple participants can be combined to build a spatiotemporal view of the phenomenon of interest and also to extract important community statistics. Given the ubiquity of mobile phones and the high density of people in metropolitan areas, participatory sensing can achieve an unprecedented level of coverage in both space and time for observing events of interest in urban spaces. Several exciting participatory sensing applications have emerged in recent years. For example, GPS traces uploaded by drivers and passengers can be used to generate real time traffic statistics. Similarly, street-level audio samples collected by pedestrians can be aggregated to create a citywide noise map. In this advanced seminar, we will provide a comprehensive overview of this new and exciting paradigm and outline the major research challenges.