Modern recommendation systems can be broadly divided into two key stages: the ranking stage, where the system predicts various user engagements (e.g., click-through rate, like rate, follow rate, watch time), and the value model stage, which aggregates these predictive scores through a function (e.g., a linear combination defined by a weight vector) to measure the value of each content by a single numerical score. Both stages play roughly equally important roles in real industrial systems; however, how to optimize the model weights for the second stage still lacks systematic study. This paper focuses on optimizing the second stage through auto-tuning technology. Although general auto-tuning systems and solutions - both from established production practices and open-source solutions - can address this problem, they typically require weeks or even months to identify a feasible solution. Such prolonged tuning processes are unacceptable in production environments for recommendation systems, as suboptimal value models can severely degrade user experience. An effective auto-tuning solution is required to identify a viable model within 2-3 days, rather than the extended timelines typically associated with existing approaches. In this paper, we introduce a practical auto-tuning system named HyperZero that addresses these time constraints while effectively solving the unique challenges inherent in modern recommendation systems. Moreover, this framework has the potential to be expanded to broader tuning tasks within recommendation systems.

Operating systems include many heuristic algorithms designed to improve overall storage performance and throughput. Because such heuristics cannot work well for all conditions and workloads, system designers resorted to exposing numerous tunable parameters to users—thus burdening users with continually optimizing their own storage systems and applications. Storage systems are usually responsible for most latency in I/O-heavy applications, so even a small latency improvement can be significant. Machine learning (ML) techniques promise to learn patterns, generalize from them, and enable optimal solutions that adapt to changing workloads. We propose that ML solutions become a first-class component in OSs and replace manual heuristics to optimize storage systems dynamically. In this article, we describe our proposed ML architecture, called KML. We developed a prototype KML architecture and applied it to two case studies: optimizing readahead and NFS read-size values. Our experiments show that KML consumes less than 4 KB of dynamic kernel memory, has a CPU overhead smaller than 0.2%, and yet can learn patterns and improve I/O throughput by as much as 2.3× and 15× for two case studies—even for complex, never-seen-before, concurrently running mixed workloads on different storage devices.

Video accessibility is crucial for blind screen-reader users as online videos are increasingly playing an essential role in education, employment, and entertainment. While there exist quite a few techniques and guidelines that focus on creating accessible videos, there is a dearth of research that attempts to characterize the accessibility of existing videos. Therefore in this paper, we define and investigate a diverse set of video and audio-based accessibility features in an effort to characterize accessible and inaccessible videos. As a ground truth for our investigation, we built a custom dataset of 600 videos, in which each video was assigned an accessibility score based on the number of its wins in a Swiss-system tournament, where human annotators performed pairwise accessibility comparisons of videos. In contrast to existing accessibility research where the assessments are typically done by blind users, we recruited sighted users for our effort, since videos comprise a special case where sight could be required to better judge if any particular scene in a video is presently accessible or not. Subsequently, by examining the extent of association between the accessibility features and the accessibility scores, we could determine the features that significantly (positively or negatively) impact video accessibility and therefore serve as good indicators for assessing the accessibility of videos. Using the custom dataset, we also trained machine learning models that leveraged our handcrafted features to either classify an arbitrary video as accessible/inaccessible or predict an accessibility score for the video. Evaluation of our models yielded an F1 score of 0.675 for binary classification and a mean absolute error of 0.53 for score prediction, thereby demonstrating their potential in video accessibility assessment while also illuminating their current limitations and the need for further research in this area.

Storage systems and their OS components are designed to accommodate a wide variety of applications and dynamic workloads. Storage components inside the OS contain various heuristic algorithms to provide high performance and adaptability for different workloads. These heuristics may be tunable via parameters, and some system calls allow users to optimize their system performance. These parameters are often predetermined based on experiments with limited applications and hardware. Thus, storage systems often run with these predetermined and possibly suboptimal values. Tuning these parameters manually is impractical: one needs an adaptive, intelligent system to handle dynamic and complex workloads. Machine learning (ML) techniques are capable of recognizing patterns, abstracting them, and making predictions on new data. ML can be a key component to optimize and adapt storage systems. In this position paper, we propose KML, an ML framework for storage systems. We implemented a prototype and demonstrated its capabilities on the well-known problem of tuning optimal readahead values. Our results show that KML has a small memory footprint, introduces negligible overhead, and yet enhances throughput by as much as 2.3x.

Gliding a finger on touchscreen to reach a target, that is, touch exploration, is a common selection method of blind screen-reader users. This paper investigates their gliding behavior and presents a model for their motor performance. We discovered that the gliding trajectories of blind people are a mixture of two strategies: 1) ballistic movements with iterative corrections relying on non-visual feedback, and 2) multiple sub-movements separated by stops, and concatenated until the target is reached. Based on this finding, we propose the mixture pointing model, a model that relates movement time to distance and width of the target. The model outperforms extant models, improving R2 from 0.65 for Fitts’ law to 0.76, and is superior in cross-validation and information criteria. The model advances understanding of gliding-based target selection and serves as a tool for designing interface layouts for screen-reader based touch exploration.

Filling out PDF forms with screen readers has always been a challenge for people who are blind. Many of these forms are not interactive and hence are not accessible; even if they are interactive, the serial reading order of the screen reader makes it difficult to associate the correct labels with the form fields. This demo will present TransPAc[5], an assistive technology that enables blind people to fill out PDF forms. Since blind people are familiar with web browsing, TransPAc leverages this fact by faithfully transforming a PDF document with forms into a HTML page. The blind user fills out the form fields in the HTML page with their screen reader and these filled-in data values are transparently transferred onto the corresponding form fields in the PDF document. TransPAc thus addresses a long standing problem in PDF form accessibility.

People with low vision who use screen magnifiers to interact with computing devices find it very challenging to interact with dynamically changing digital content such as videos, since they do not have the luxury of time to manually move, i.e., pan the magnifier lens to different regions of interest (ROIs) or zoom into these ROIs before the content changes across frames. In this paper, we present SViM, a first of its kind screen-magnifier interface for such users that leverages advances in computer vision, particularly video saliency models, to identify salient ROIs in videos. SViM's interface allows users to zoom in/out of any point of interest, switch between ROIs via mouse clicks and provides assistive panning with the added flexibility that lets the user explore other regions of the video besides the ROIs identified by SViM. Subjective and objective evaluation of a user study with 13 low vision screen magnifier users revealed that overall the participants had a better user experience with SViM over extant screen magnifiers, indicative of the former's promise and potential for making videos accessible to low vision screen magnifier users.

Navigating webpages with screen readers is a challenge even with recent improvements in screen reader technologies and the increased adoption of web standards for accessibility, namely ARIA. ARIA landmarks, an important aspect of ARIA, lets screen reader users access different sections of the webpage quickly, by enabling them to skip over blocks of irrelevant or redundant content. However, these landmarks are sporadically and inconsistently used by web developers, and in many cases, even absent in numerous web pages. Therefore, we propose SaIL, a scalable approach that automatically detects the important sections of a web page, and then injects ARIA landmarks into the corresponding HTML markup to facilitate quick access to these sections. The central concept underlying SaIL is visual saliency, which is determined using a state-of-the-art deep learning model that was trained on gaze-tracking data collected from sighted users in the context of web browsing. We present the findings of a pilot study that demonstrated the potential of SaIL in reducing both the time and effort spent in navigating webpages with screen readers.

PDF forms are ubiquitous. Businesses big and small, government agencies, health and educational institutions and many others have all embraced PDF forms. People use PDF forms for providing information to these entities. But people who are blind frequently find it very difficult to fill out PDF forms with screen readers, the standard assistive software that they use for interacting with computer applications. Firstly, many of the them are not even accessible as they are non-interactive and hence not editable on a computer. Secondly, even if they are interactive, it is not always easy to associate the correct labels with the form fields, either because the labels are not meaningful or the sequential reading order of the screen reader misses the visual cues that associate the correct labels with the fields. In this paper we present a solution to the accessibility problem of PDF forms. We leverage the fact that many people with visual impairments are familiar with web browsing and are proficient at filling out web forms. Thus, we create a web form layer over the PDF form via a high fidelity transformation process that attempts to preserve all the spatial relationships of the PDF elements including forms, their labels and the textual content. Blind people only interact with the web forms, and the filled out web form fields are transparently transferred to the corresponding fields in the PDF form. An optimization algorithm automatically adjusts the length and width of the PDF fields to accommodate arbitrary size field data. This ensures that the filled out PDF document does not have any truncated form-field values, and additionally, it is readable. A user study with fourteen users with visual impairments revealed that they were able to populate more form fields than the status quo and the self-reported user experience with the proposed interface was superior compared to the status quo.

Predicting promising academic papers is useful for a variety of parties, including researchers, universities, scientific councils, and policymakers. Researchers may benefit from such data to narrow down their reading list and focus on what will be important, and policymakers may use predictions to infer rising fields for a more strategic distribution of resources. This paper proposes a novel technique to predict a paper's future impact (i.e., number of citations) by using temporal and topological features derived from citation networks. We use a behavioral modeling approach in which the temporal change in the number of citations a paper gets is clustered, and new papers are evaluated accordingly. Then, within each cluster, we model the impact prediction as a regression problem where the objective is to predict the number of citations that a paper will get in the near or far future, given the early citation performance of the paper. The results of empirical evaluations on data from several well-known citation databases show that the proposed framework performs significantly better than the state of the art approaches.