Air pollution, particularly particulate matter 2.5 (PM2.5), poses a significant public health challenge in densely populated developing regions. Moreover, deploying an extensive ground sensor network to monitor PM2.5accurately is economically unfeasible in such regions. To address this problem, we utilize Sparse Variational Gaussian Process (SVGP) models to generate approximate data using the limited ground sensor data. Since SVGPs use computational approximators for Gaussian Process modeling, we hypothesize that their inducing points can be trained to adapt spatially, i.e., these points, when optimized, can spread over the region of interest. Hence, well-initialized inducing points allow SVGPs to model PM2.5data by capturing spatial variations of the region. We evaluate our hypothesis using PM2.5data from Lima, Peru, one of the most polluted cities in the Americas, and with very few PM2.5ground sensors. Our experiments qualitatively validate our hypothesis of spatial adaptation and provide a quantitative justification of improved performance over the baseline models.

In this paper, we tackle the complex task of analyzing televised debates, with a focus on a prime time news debate show from India. Previous methods, which often relied solely on text, fall short in capturing the multimodal essence of these debates [27]. To address this gap, we introduce a comprehensive automated toolkit that employs advanced computer vision and speech-to-text techniques for large-scale multimedia analysis. Utilizing state-of-the-art computer vision algorithms and speech-to-text methods, we transcribe, diarize, and analyze thousands of YouTube videos of a prime-time television debate show in India. These debates are a central part of Indian media but have been criticized for compromised journalistic integrity and excessive dramatization [18]. Our toolkit provides concrete metrics to assess bias and incivility, capturing a comprehensive multimedia perspective that includes text, audio utterances, and video frames. Our findings reveal significant biases in topic selection and panelist representation, along with alarming levels of incivility. This work offers a scalable, automated approach for future research in multimedia analysis, with profound implications for the quality of public discourse and democratic debate. To catalyze further research in this area, we also release the code, dataset collected and supplemental pdf.1

With the widespread increase in hateful content on the web, hate detection has become more crucial than ever. Although vast literature exists on hate detection from text, images and videos, interestingly, there has been no previous work on hateful comment detection (HCD) from video pages. HCD is critical for comment moderation and for flagging controversial videos. Comments are often short, contextual and convoluted making the problem challenging. Toward solving this problem, we contribute a dataset, HateComments, consisting of 2071 comments for 401 videos obtained from two popular video sharing platforms. We investigate two related tasks: binary HCD and 4-class multi-label hate target-type prediction (HTP). We systematically explore the importance of various forms of context for effective HCD. Our initial experiments show that our best method which leverages rich video context (like description, transcript and visual input) leads to an HCD accuracy of ~78.6% and an ROC AUC score of ~0.61 for HTP. Code and data is at https://drive.google.com/file/d/1EUbWDUokv1CYkWKlwByUC6yIuBGUw2MN/.

Abstract When individuals encounter new information (data), that information is incorporated with their existing beliefs (prior) to form a new belief (posterior) in a process referred to as belief updating. While most studies on rational belief updating in visual data analysis elicit beliefs immediately after data is shown, we posit that there may be critical movement in an individual's beliefs when elicited immediately after data is shown v. after a temporal delay (e.g., due to forgetfulness or weak incorporation of the data). Our paper investigates the hypothesis that posterior beliefs elicited after a time interval will “decay” back towards the prior beliefs compared to the posterior beliefs elicited immediately after new data is presented. In this study, we recruit 101 participants to complete three tasks where beliefs are elicited immediately after seeing new data and again after a brief distractor task. We conduct (1) a quantitative analysis of the results to understand if there are any systematic differences in beliefs elicited immediately after seeing new data or after a distractor task and (2) a qualitative analysis of participants' reflections on the reasons for their belief update. While we find no statistically significant global trends across the participants beliefs elicited immediately v. after the delay, the qualitative analysis provides rich insight into the reasons for an individual's belief movement across 9 prototypical scenarios, which includes (i) decay of beliefs as a result of either forgetting the information shown or strongly held prior beliefs, (ii) strengthening of confidence in updated beliefs by positively integrating the new data and (iii) maintaining a consistently updated belief over time, among others. These results can guide subsequent experiments to disambiguate when and by what mechanism new data is truly incorporated into one's belief system.

In the last two decades, Information and Communication Technologies (ICTs) have played a pivotal role in empowering rural populations in India by making knowledge more accessible. Digital Green is one such ICT that employs a participatory approach with smallholder farmers to produce instructional agricultural videos that encompass content specific to them. With the help of human mediators, they disseminate these videos to farmers using projectors to improve the adoption of agricultural practices. Digital Green’s web-based data tracker (CoCo) stores the attendance and adoption logs of millions of farmers, the videos screened to them and their demographic information. In our work, we leverage this data for a period of ten years between 2010-2020 across five states in India where Digital Green is most active and use it to conduct a holistic evaluation of the ICT. First, we find disparities in the adoption rates of farmers, following which we use statistical tests to identify the different factors that lead to these disparities as well as gender-based inequalities. We find that farmers with higher adoption rates adopt videos of shorter duration and belong to smaller villages. Second, to provide assistance to farmers facing challenges, we model the adoption of practices from a video as a prediction problem and experiment with different model architectures. Our classifier achieves accuracies ranging from 79% to 90% across the five states, demonstrating its potential for assisting future ethnographic investigations. Third, we use SHAP values in conjunction with our model for explaining the impact of various network, content and demographic features on adoption. Our research finds that farmers greatly benefit from past adopters of a video from their group and village. We also discover that videos with a low content-specificity benefit some farmers more than others. Next, we highlight the implications of our findings by translating them into recommendations for providing focused assistance, community building, video screening, revisiting participatory approach and mitigating inequalities. Lastly, we conclude with a discussion on how our work can assist future investigations into the lived experiences of farmers.

Movement data collected through GPS or other technologies is increasingly common, but is difficult to visualize due to overplotting and occlusion of movements when displayed on 2D maps. An additional challenge is the extraction of useful higher-level information (such as meetings) derived from the raw movement data. We present a design study of MovementSlicer, a tool for visualizing the places visited, and behaviors of, individual actors, and also the meetings between multiple actors. We first present a taxonomy of visualizations of movement data, and then consider tasks to support when analyzing movement data and especially meetings of multiple actors. We argue that Gantt charts have many advantages for understanding the movements and meetings of small groups of moving entities, and present the design of a Gantt chart that can nest people within locations or locations within people along the vertical axis, and show time along the horizontal axis. The rows of our Gantt chart are sorted by activity level and can be filtered using a weighted adjacency matrix showing meetings between people. Empty time intervals in the Gantt chart can be automatically folded, with smoothly animated transitions, yielding a multi-focal view. Case studies demonstrate the utility of our prototype.

We design and evaluate a multitouch radial menu for large screens with two desirable properties. First, it allows a single gesture to select a command and then continuously control arguments for that command with unbroken kinesthetic tension. Second, arguments are controlled with 1 or 2 fingers for up to 4 degrees of freedom (DoF). For example, the user may select one command for 4 DoF direct manipulation (translation + scaling + rotation), or another command for 3 DoF camera operations (pan + zoom), using the same two-finger pinch gesture, but with different initial orientations of the gesture to disambiguate. We present a taxonomy to classify previous menuing techniques sharing the first property, and discuss how very few techniques have both of these properties. Our work also extends previous work by Banovic et al. in the following ways: our menu supports submenus and a fast default command, and we experimentally evaluate the effect of varying the number of rings in the menu, the symmetry of the menu, and the use of one hand vs. two hands vs. a stylus and hand.

We study a sensor node with an energy harvesting source. The generated energy can be stored in a buffer. The sensor node periodically senses a random field and generates a packet. These packets are stored in a queue and transmitted using the energy available at that time. We obtain energy management policies that are throughput optimal, i.e., the data queue stays stable for the largest possible data rate. Next we obtain energy management policies which minimize the mean delay in the queue. We also compare performance of several easily implementable sub-optimal energy management policies. A greedy policy is identified which, in low SNR regime, is throughput optimal and also minimizes mean delay.