Contrastive learning (CL) has emerged as a promising approach for representation learning in time series data by embedding similar pairs closely while distancing dissimilar ones. However, existing CL methods often introduce false negative pairs (FNPs) by neglecting inherent characteristics and then randomly selecting distinct segments as dissimilar pairs, leading to erroneous representation learning, reduced model performance, and overall inefficiency. To address these issues, we systematically define and categorize FNPs in time series into semantic false negative pairs and temporal false negative pairs for the first time: the former arising from overlooking similarities in label categories, which correlates with similarities in non-stationarity and the latter from neglecting temporal proximity. Moreover, we introduce StatioCL, a novel CL framework that captures non-stationarity and temporal dependency to mitigate both FNPs and rectify the inaccuracies in learned representations. By interpreting and differentiating non-stationary states, which reflect the correlation between trends or temporal dynamics with underlying data patterns, StatioCL effectively captures the semantic characteristics and eliminates semantic FNPs. Simultaneously, StatioCL establishes fine-grained similarity levels based on temporal dependencies to capture varying temporal proximity between segments and to mitigate temporal FNPs. Evaluated on real-world benchmark time series classification datasets, StatioCL demonstrates a substantial improvement over state-of-the-art CL methods, achieving a 2.9% increase in Recall and a 19.2% reduction in FNPs. Most importantly, StatioCL also shows enhanced data efficiency and robustness against label scarcity.

Federated Learning (FL) enables model development by leveraging data distributed across numerous edge devices without transferring local data to a central server. However, existing FL methods still face challenges when dealing with scarce and label-skewed data across devices, resulting in local model overfitting and drift, consequently hindering the performance of the global model. In response to these challenges, we propose a pioneering framework called FLea, incorporating the following key components: i) A global feature buffer that stores activation-target pairs shared from multiple clients to support local training. This design mitigates local model drift caused by the absence of certain classes; ii) A feature augmentation approach based on local and global activation mix-ups for local training. This strategy enlarges the training samples, thereby reducing the risk of local overfitting; iii) An obfuscation method to minimize the correlation between intermediate activations and the source data, enhancing the privacy of shared features. To verify the superiority of FLea, we conduct extensive experiments using a wide range of data modalities, simulating different levels of local data scarcity and label skew. The results demonstrate that FLea consistently outperforms state-of-the-art FL counterparts (among 13 of the experimented 18 settings, the improvement is over 5%) while concurrently mitigating the privacy vulnerabilities associated with shared features.

Time series forecasting, as one of the fundamental machine learning areas, has attracted tremendous attentions over recent years. The solutions have evolved from statistical machine learning (ML) methods to deep learning techniques. One emerging sub-field of time series forecasting is individual disease progression forecasting, e.g., predicting individuals' disease development over a few days (e.g., deteriorating trends, recovery speed) based on few past observations. Despite the promises in the existing ML techniques, a variety of unique challenges emerge for disease progression forecasting, such as irregularly-sampled time series, data sparsity, and individual heterogeneity in disease progression. To tackle these challenges, we propose novel Conditional Neural Ordinary Differential Equations Processes (CNDPs), and validate it in a COVID-19 disease progression forecasting task using audio data. CNDPs allow for irregularly-sampled time series modelling, enable accurate forecasting with sparse past observations, and achieve individual-level progression forecasting. CNDPs show strong performance with an Unweighted Average Recall (UAR) of 78.1%, outperforming a variety of commonly used Recurrent Neural Networks based models. With the proposed label-enhancing mechanism (i.e., including the initial health status as input) and the customised individual-level loss, CNDPs further boost the performance reaching a UAR of 93.6%. Additional analysis also reveals the model's capability in tracking individual-specific recovery trend, implying the potential usage of the model for remote disease progression monitoring. In general, CNDPs pave new pathways for time series forecasting, and provide considerable advantages for disease progression monitoring.

Modern machine learning applications should be able to address the intrinsic challenges arising over inference on massive real-world datasets, including scalability and robustness to outliers. Despite the multiple benefits of Bayesian methods (such as uncertainty-aware predictions, incorporation of experts knowledge, and hierarchical modeling), the quality of classic Bayesian inference depends critically on whether observations conform with the assumed data generating model, which is impossible to guarantee in practice. In this work, we propose a variational inference method that, in a principled way, can simultaneously scale to large datasets, and robustify the inferred posterior with respect to the existence of outliers in the observed data. Reformulating Bayes theorem via the β-divergence, we posit a robustified generalized Bayesian posterior as the target of inference. Moreover, relying on the recent formulations of Riemannian coresets for scalable Bayesian inference, we propose a sparse variational approximation of the robustified posterior and an efficient stochastic black-box algorithm to construct it. Overall our method allows releasing cleansed data summaries that can be applied broadly in scenarios involving structured and unstructured data contamination. We illustrate the applicability of our approach in diverse simulated and real datasets, and various statistical models, including Gaussian mean inference, logistic and neural linear regression, demonstrating its superiority to existing Bayesian summarization methods in the presence of outliers.

Audio signals generated by the human body (e.g., sighs, breathing, heart, digestion, vibration sounds) have routinely been used by clinicians as indicators to diagnose disease or assess disease progression. Until recently, such signals were usually collected through manual auscultation at scheduled visits. Research has now started to use digital technology to gather bodily sounds (e.g., from digital stethoscopes) for cardiovascular or respiratory examination, which could then be used for automatic analysis. Some initial work shows promise in detecting diagnostic signals of COVID-19 from voice and coughs. In this paper we describe our data analysis over a large-scale crowdsourced dataset of respiratory sounds collected to aid diagnosis of COVID-19. We use coughs and breathing to understand how discernible COVID-19 sounds are from those in asthma or healthy controls. Our results show that even a simple binary machine learning classifier is able to classify correctly healthy and COVID-19 sounds. We also show how we distinguish a user who tested positive for COVID-19 and has a cough from a healthy user with a cough, and users who tested positive for COVID-19 and have a cough from users with asthma and a cough. Our models achieve an AUC of above 80% across all tasks. These results are preliminary and only scratch the surface of the potential of this type of data and audio-based machine learning. This work opens the door to further investigation of how automatically analysed respiratory patterns could be used as pre-screening signals to aid COVID-19 diagnosis.

Smartphones have started to be used as self reporting tools for mental health state as they accompany individuals during their days and can therefore gather temporally fine grained data. However, the analysis of self reported mood data offers challenges related to non-homogeneity of mood assessment among individuals due to the complexity of the feeling and the reporting scales, as well as the noise and sparseness of the reports when collected in the wild. In this paper, we propose a new end-to-end ML model inspired by video frame prediction and machine translation, that forecasts future sequences of mood from previous self-reported moods collected in the real world using mobile devices. Contrary to traditional time series forecasting algorithms, our multi-task encoder-decoder recurrent neural network learns patterns from different users, allowing and improving the prediction for users with limited number of self-reports. Unlike traditional feature-based machine learning algorithms, the encoder-decoder architecture enables to forecast a sequence of future moods rather than one single step. Meanwhile, multi-task learning exploits some unique characteristics of the data (mood is bi-dimensional), achieving better results than when training single-task networks or other classifiers.

Point-of-Interest (POI) recommender systems play a vital role in people's lives by recommending unexplored POIs to users and have drawn extensive attention from both academia and industry. Despite their value, however, they still suffer from the challenges of capturing complicated user preferences and fine-grained user-POI relationship for spatio-temporal sensitive POI recommendation. Existing recommendation algorithms, including both shallow and deep approaches, usually embed the visiting records of a user into a single latent vector to model user preferences: this has limited power of representation and interpretability. In this paper, we propose a novel topic-enhanced memory network (TEMN), a deep architecture to integrate the topic model and memory network capitalising on the strengths of both the global structure of latent patterns and local neighbourhood-based features in a nonlinear fashion. We further incorporate a geographical module to exploit user-specific spatial preference and POI-specific spatial influence to enhance recommendations. The proposed unified hybrid model is widely applicable to various POI recommendation scenarios. Extensive experiments on real-world WeChat datasets demonstrate its effectiveness (improvement ratio of 3.25% and 29.95% for context-aware and sequential recommendation, respectively). Also, qualitative analysis of the attention weights and topic modeling provides insight into the model's recommendation process and results.

The investigation of the effect of the built environment in a neighbourhood and how it impacts residents' health is of value to researchers from public health policy to social science. The traditional methods to assess this impact is through surveys which lead to temporally and spatially coarse grained data and are often not cost effective. Here we propose an approach to link the effects of neighbourhood services over citizen health using a technique that attempts to highlight the cause-effect aspects of these relationships. The method is based on the theory of propensity score matching with multiple 'doses' and it leverages existing fine grained open web data. To demonstrate the method, we study the effect of sport venue presence on the prevalence of antidepressant prescriptions in over 600 neighbourhoods in London over a period of three years. We find the distribution of effects is approximately normal, centred on a small negative effect on prescriptions with increases in the availability of sporting facilities, on average. We assess the procedure through some standard quantitative metrics as well as matching on synthetic data generated by modelling the real data. This approach opens the door to fast and inexpensive alternatives to quantify and continuously monitor effects of the neighborhood built environment on population health.

As modern transportation systems become more complex, there is need for mobile applications that allow travelers to navigate efficiently in cities. In taxi transport the recent proliferation of Uber has introduced new norms including a flexible pricing scheme where journey costs can change rapidly depending on passenger demand and driver supply. To make informed choices on the most appropriate provider for their journeys, travelers need access to knowledge about provider pricing in real time. To this end, we developed OpenStreetCab a mobile application that offers advice on taxi transport comparing provider prices. We describe its development and deployment in two cities, London and New York, and analyse thousands of user journey queries to compare the price patterns of Uber against major local taxi providers. We have observed large heterogeneity across the taxi transport markets in the two cities. This motivated us to perform a price validation and measurement experiment on the ground comparing Uber and Black Cabs in London. The experimental results reveal interesting insights: not only they confirm feedback on pricing and service quality received by professional driver users, but also they reveal the tradeoffs between prices and journey times between taxi providers. With respect to journey times in particular, we show how experienced taxi drivers, in the majority of the cases, are able to navigate faster to a destination compared to drivers who rely on modern navigation systems. We provide evidence that this advantage becomes stronger in the centre of a city where urban density is high.

Cultural activity is an inherent aspect of urban life and the success of a modern city is largely determined by its capacity to offer generous cultural entertainment to its citizens. To this end, the optimal allocation of cultural establishments and related resources across urban regions becomes of vital importance, as it can reduce financial costs in terms of planning and improve quality of life in the city, more generally. In this paper, we make use of a large longitudinal dataset of user location check-ins from the online social network WeChat to develop a data-driven framework for cultural planning in the city of Beijing. We exploit rich spatio-temporal representations on user activity at cultural venues and use a novel extended version of the traditional latent Dirichlet allocation model that incorporates temporal information to identify latent patterns of urban cultural interactions. Using the characteristic typologies of mobile user cultural activities emitted by the model, we determine the levels of demand for different types of cultural resources across urban areas. We then compare those with the corresponding levels of supply as driven by the presence and spatial reach of cultural venues in local areas to obtain high resolution maps that indicate urban regions with lack of cultural resources, and thus give suggestions for further urban cultural planning and investment optimisation.

Estimating revenue and business demand of a newly opened venue is paramount as these early stages often involve critical decisions such as first rounds of staffing and resource allocation. Traditionally, this estimation has been performed through coarse measures such as observing numbers in local venues. The advent of crowdsourced data from devices and services has opened the door to better predictions of temporal visitation patterns for locations and venues. In this paper, using mobility data from the location-based service Foursquare, we treat venue categories as proxies for urban activities and analyze how they become popular over time. The main contribution of this work is a prediction framework able to use characteristic temporal signatures of places together with k-nearest neighbor metrics capturing similarities among urban regions to forecast weekly popularity dynamics of a new venue establishment. Our evaluation shows that temporally similar areas of a city can be valuable predictors, decreasing error by 41%. Our findings have the potential to impact the design of location-based technologies and decisions made by new business owners.

Measuring mental well-being with mobile sensing has been an increasingly active research topic. Pervasiveness of smartphones combined with the convenience of mobile app distribution platforms (e.g., Google Play) provide a tremendous opportunity to reach out to millions of users. However, the studies at the confluence of mental health and mobile sensing have been longitudinally limited, controlled, or confined to a small number of participants. In this paper we report on what we believe is the largest longitudinal in-the-wild study of mood through smartphones. We describe an Android app to collect participants' self-reported moods and system triggered experience sampling data while passively measuring their physical activity, sociability, and mobility via their device's sensors. We report the results of a large-scale analysis of the data collected for about three years from 18,000 users.

Summary form only given. With the advent of powerful and inexpensive sensing technology the ability to study human behaviour and activity at large scale and for long periods is becoming a firm reality. Wearables and mobile devices further allow the continuous physical colocation with the users. This reality generates new challenges but also opens the door to potentially innovative ways of understanding our daily lives. In this talk we will discuss our experience in large mobile sensor deployments. We will discuss the issues raised by mobile sensing big data in terms of data crowdsourcing, continuous sensing challenges, data analysis, privacy, user feedback. Examples will be drawn from our healthcare, transport, urban planning and organization analytics studies.

Large metropolitan cities bring together diverse individuals, creating opportunities for cultural and intellectual exchanges, which can ultimately lead to social and economic enrichment. In this work, we present a novel network perspective on the interconnected nature of people and places, allowing us to capture the social diversity of urban locations through the social network and mobility patterns of their visitors. We use a dataset of approximately 37K users and 42K venues in London to build a network of Foursquare places and the parallel Twitter social network of visitors through check-ins. We define four metrics of the social diversity of places which relate to their social brokerage role, their entropy, the homogeneity of their visitors and the amount of serendipitous encounters they are able to induce. This allows us to distinguish between places that bring together strangers versus those which tend to bring together friends, as well as places that attract diverse individuals as opposed to those which attract regulars. We correlate these properties with wellbeing indicators for London neighbourhoods and discover signals of gentrification in deprived areas with high entropy and brokerage, where an influx of more affluent and diverse visitors points to an overall improvement of their rank according to the UK Index of Multiple Deprivation for the area over the five-year census period. Our analysis sheds light on the relationship between the prosperity of people and places, distinguishing between different categories and urban geographies of consequence to the development of urban policy and the next generation of socially-aware location-based applications.

The problem of identifying the optimal location for a new retail store has been the focus of past research, especially in the field of land economy, due to its importance in the success of a business. Traditional approaches to the problem have factored in demographics, revenue and aggregated human flow statistics from nearby or remote areas. However, the acquisition of relevant data is usually expensive. With the growth of location-based social networks, fine grained data describing user mobility and popularity of places has recently become attainable.

Inferring the type of activities in neighborhoods of urban centers may be helpful in a number of contexts including urban planning, content delivery and activity recommendations for mobile web users or may even yield to a deeper understanding of the geographical evolution of social life in the city . During the past few years, the analysis of mobile phone usage patterns, or of social media with longitudinal attributes, have aided the automatic characterization of the dynamics of the urban environment. In this work, we combine a dataset sourced from a telecommunication provider in Spain with a database of millions of geotagged venues from Foursquare and we formulate the problem of urban activity inference in a supervised learning framework. In particular, we exploit user communication patterns observed at the base station level in order to predict the activity of Foursquare users who checkin-in at nearby venues. First, we mine a set of machine learning features that allow us to encode the input telecommunication signal of a tower. Subsequently, we evaluate a diverse set of supervised learning algorithms using labels extracted from Foursquare place categories and we consider two application scenarios. Initially, we assess how hard it is to predict specific urban activity of an area, showing that Nightlife and Entertainment spots are those easier to infer, whereas College and Shopping areas are those featuring the lowest accuracy rates. Then, considering a candidate set of activity types in a geographic area, we aim to elect the most prominent one. We demonstrate how the difficulty of the problem increases with the number of classes incorporated in the prediction task, yet the classifiers achieve a considerably better performance compared to a random guess even when the set of candidate classes increases.

Mobile location-based services are thriving, providing an unprecedented opportunity to collect fine grained spatio-temporal data about the places users visit. This multi-dimensional source of data offers new possibilities to tackle established research problems on human mobility, but it also opens avenues for the development of novel mobile applications and services. In this work we study the problem of predicting the next venue a mobile user will visit, by exploring the predictive power offered by different facets of user behavior. We first analyze about 35 million check-ins made by about 1 million Foursquare users in over 5 million venues across the globe, spanning a period of five months. We then propose a set of features that aim to capture the factors that may drive users' movements. Our features exploit information on transitions between types of places, mobility flows between venues, and spatio-temporal characteristics of user check-in patterns. We further extend our study combining all individual features in two supervised learning models, based on linear regression and M5 model trees, resulting in a higher overall prediction accuracy. We find that the supervised methodology based on the combination of multiple features offers the highest levels of prediction accuracy: M5 model trees are able to rank in the top fifty venues one in two user check-ins, amongst thousands of candidate items in the prediction list.

Link prediction systems have been largely adopted to recommend new friends in online social networks using data about social interactions. With the soaring adoption of location-based social services it becomes possible to take advantage of an additional source of information: the places people visit. In this paper we study the problem of designing a link prediction system for online location-based social networks. We have gathered extensive data about one of these services, Gowalla, with periodic snapshots to capture its temporal evolution. We study the link prediction space, finding that about 30% of new links are added among "place-friends", i.e., among users who visit the same places. We show how this prediction space can be made 15 times smaller, while still 66% of future connections can be discovered. Thus, we define new prediction features based on the properties of the places visited by users which are able to discriminate potential future links among them.

Providers such as YouTube offer easy access to multimedia content to millions, generating high bandwidth and storage demand on the Content Delivery Networks they rely upon. More and more, the diffusion of this content happens on online social networks such as Facebook and Twitter, where social cascades can be observed when users increasingly repost links they have received from others. In this paper we describe how geographic information extracted from social cascades can be exploited to improve caching of multimedia files in a Content Delivery Network. We take advantage of the fact that social cascades can propagate in a geographically limited area to discern whether an item is spreading locally or globally. This informs cache replacement policies, which utilize this information to ensure that content relevant to a cascade is kept close to the users who may be interested in it. We validate our approach by using a novel dataset which combines social interaction data with geographic information: we track social cascades of YouTube links over Twitter and build a proof-of-concept geographic model of a realistic distributed Content Delivery Network. Our performance evaluation shows that we are able to improve cache hits with respect to cache policies without geographic and social information.

The use of the World Wide Web as a communication medium for knowledge engineers and software designers is limited by the lack of tools for writing, sharing, and verifying documents written with design notations. For instance, the Z language has a rich set of mathematical characters, and requires graphic-rich boxes and schemas for structuring a specification document. It is difficult to integrate Z specifications and text on WWW pages written with HTML, and traditional tools are not suited for the task. On the other hand, a newly proposed standard for markup languages, namely XML, allows one to define any set of markup elements; hence, it is suitable for describing any kind of notation. Unfortunately, the proposed standard for rendering XML documents, namely XSL, provides for text-only (although sophisticated) rendering of XML documents, and thus it cannot be used for more complex notations. We present a Java-based tool for applying any notation to elements of XML documents. These XML documents can thus be shown on current-generation WWW browsers with Java capabilities. A complete package for displaying Z specifications has been implemented and integrated with standard text parts. Being a complete rendering engine, text parts and Z specifications can be freely intermixed, and all the standard features of XML (including HTML links and form elements) are available outside and inside Z specifications. Furthermore, the extensibility of our engine allows any additional notations to be supported and integrated with the ones we describe.