Search for: All records

Transfer learning leverages feature representations of deep neural networks (DNNs) pretrained on source tasks with rich data to empower effective finetuning on downstream tasks. However, the pre-trained models are often prohibitively large for delivering generalizable representations, which limits their deployment on edge devices with constrained resources. To close this gap, we propose a new transfer learning pipeline, which leverages our finding that robust tickets can transfer better, i.e., subnetworks drawn with properly induced adversarial robustness can win better transferability over vanilla lottery ticket subnetworks. Extensive experiments and ablation studies validate that our proposed transfer learning pipeline can achieve enhanced accuracy-sparsity trade-offs across both diverse downstream tasks and sparsity patterns, further enriching the lottery ticket hypothesis. 

Siblings play a crucial and long-lasting role in family connections and relationships. However, with the older sibling transitioning out of their parental home, maintaining a close sibling relationship can be challenging, especially if siblings have a large age difference. We conducted a diary and interview study with nine families in China which have spaced siblings, to identify design opportunities for technology to better support their communication and connection needs. We contribute to the HCI community in three aspects. First, we contribute an empirical understanding of current communication patterns from distributed families with large age gap siblings in China. Second, we identify current facilitation roles, practices, and challenges regarding sibling relationships from different stakeholders’ perspectives. Last but not least, we present technological opportunities for supporting the large-gap sibling relationship, informing directions for future research and design for distributed families. 

Studies comparing single cell RNA-Seq (scRNA-Seq) data between conditions mainly focus on differences in the proportion of cell types or on differentially expressed genes. In many cases these differences are driven by changes in cell interactions which are challenging to infer without spatial information. To determine cell-cell interactions that differ between conditions we developed the Cell Interaction Network Inference (CINS) pipeline. CINS combines Bayesian network analysis with regression-based modeling to identify differential cell type interactions and the proteins that underlie them. We tested CINS on a disease case control and on an aging mouse dataset. In both cases CINS correctly identifies cell type interactions and the ligands involved in these interactions improving on prior methods suggested for cell interaction predictions. We performed additional mouse aging scRNA-Seq experiments which further support the interactions identified by CINS. 

Recovery from substance abuse disorders (SUDs) is a lifelong process of change. Self-tracking technologies have been proposed by the recovery community as a beneficial design space to support people adopting positive lifestyles and behaviors in their recovery. To explore the potential of this design space, we designed and deployed a technology probe consisting of a mobile app, wearable visualization, and ambient display to enable people to track and reflect on the activities they adopted in their recovery process. With this probe we conducted a four-week exploratory field study with 17 adults in early recovery to investigate 1) what activities people in recovery desire to track, 2) how people perceive self-tracking tools in relation to their recovery process, and 3) what digital resources self-tracking tools can provide to aid the recovery process. Our findings illustrate the array of activities that people track in their recovery, along with usage scenarios, preferences and design tensions that arose. We discuss implications for holistic self-tracking technologies and opportunities for future work in behavior change support for this context. 

Prior research has highlighted opportunities for technology to better support the tabletop game experience in offline and online settings, but little work has focused on the social aspect of tabletop gaming. We investigated the social and collaborative aspects of tabletop gaming in the unique context of “social distancing” during the 2020 COVID-19 pandemic to shed light on the experience of remote tabletop gaming. With a multi-method qualitative approach (including digital ethnography and in-depth interviews), we empirically studied how people appropriate existing technologies and adapt their offline practices to play tabletop games remotely. We identify three themes that describe people’s game and social experience during remote play: creating a shared tabletop environment (shared space), enabling a collective understanding (shared information and awareness), and facilitating a communal temporal experience (shared time). We reflect on challenges and design opportunities for a better experience in the age of remote collaboration. 

Swing oscillation is widely observed among indoor miniature autonomous blimps (MABs) due to their underactuated design and unique aerodynamic shape. This paper presents the modeling, identification and control system design that reduce the swing oscillation of an MAB during hovering flight. We establish a dynamic model to describe the swing motion of the MAB. The model parameters are identified from both physical measurements, computer modeling and experimental data captured during flight. A control system is designed to stabilize the swing motion with features including low latency and center-of-mass (CM) position estimation. The modeling and control methods are verified with the Georgia-Tech Miniature Autonomous Blimp (GT-MAB) during hovering flight. The experimental results show that the proposed methods can effectively reduce the swing oscillation of GT-MAB. 

We investigate the mechanical response of packings of purely repulsive, frictionless disks to quasistatic deformations. The deformations include simple shear strain at constant packing fraction and at constant pressure, “polydispersity” strain (in which we change the particle size distribution) at constant packing fraction and at constant pressure, and isotropic compression. For each deformation, we show that there are two classes of changes in the interparticle contact networks: jump changes and point changes. Jump changes occur when a contact network becomes mechanically unstable, particles “rearrange”, and the potential energy (when the strain is applied at constant packing fraction) or enthalpy (when the strain is applied at constant pressure) and all derivatives are discontinuous. During point changes, a single contact is either added to or removed from the contact network. For repulsive linear spring interactions, second- and higher-order derivatives of the potential energy/enthalpy are discontinuous at a point change, while for Hertzian interactions, third- and higher-order derivatives of the potential energy/enthalpy are discontinuous. We illustrate the importance of point changes by studying the transition from a hexagonal crystal to a disordered crystal induced by applying polydispersity strain. During this transition, the system only undergoes point changes, with no jump changes. We emphasize that one must understand point changes, as well as jump changes, to predict the mechanical properties of jammed packings. 

Artificial intelligence algorithms have been used to enhance a wide variety of products and services, including assisting human decision making in high-stake contexts. However, these algorithms are complex and have trade-offs, notably between prediction accuracy and fairness to population subgroups. This makes it hard for designers to understand algorithms and design products or services in a way that respects users' goals, values, and needs. We proposed a method to help designers and users explore algorithms, visualize their trade-offs, and select algorithms with trade-offs consistent with their goals and needs. We evaluated our method on the problem of predicting criminal defendants' likelihood to re-offend through (i) a large-scale Amazon Mechanical Turk experiment, and (ii) in-depth interviews with domain experts. Our evaluations show that our method can help designers and users of these systems better understand and navigate algorithmic trade-offs. This paper contributes a new way of providing designers the ability to understand and control the outcomes of algorithmic systems they are creating.