Hierarchical structures of motion exist across research fields, including computer vision, graphics, and robotics, where complex dynamics typically arise from coordinated interactions among simpler motion components. Existing methods to model such dynamics typically rely on manually-defined or heuristic hierarchies with fixed motion primitives, limiting their generalizability across different tasks. In this work, we propose a general hierarchical motion modeling method that learns structured, interpretable motion relationships directly from data. Our method represents observed motions using graph-based hierarchies, explicitly decomposing global absolute motions into parent-inherited patterns and local motion residuals. We formulate hierarchy inference as a differentiable graph learning problem, where vertices represent elemental motions and directed edges capture learned parent-child dependencies through graph neural networks. We evaluate our hierarchical reconstruction approach on three examples: 1D translational motion, 2D rotational motion, and dynamic 3D scene deformation via Gaussian splatting. Experimental results show that our method reconstructs the intrinsic motion hierarchy in 1D and 2D cases, and produces more realistic and interpretable deformations compared to the baseline on dynamic 3D Gaussian splatting scenes. By providing an adaptable, data-driven hierarchical modeling paradigm, our method offers a formulation applicable to a broad range of motion-centric tasks.

The ability to collect and assess different types of data from Unmanned Aerial Systems in real-time or near real-time is still a difficult challenge. Issues faced can include significant computation times, signal strength and wireless transmission delays or limitations, open source software versus proprietary data, etc. To attempt to remedy such issues, we aim for achieving on-board capabilities for automatic object detection, image analysis, and adaptable situational assessment. The result is real-time comparison between flights as well as preliminary data assessment using techniques from topological data analysis.

Autopilot systems are typically composed of an “inner loop” providing stability and control, whereas an “outer loop” is responsible for mission-level objectives, such as way-point navigation. Autopilot systems for unmanned aerial vehicles are predominately implemented using Proportional-Integral-Derivative (PID) control systems, which have demonstrated exceptional performance in stable environments. However, more sophisticated control is required to operate in unpredictable and harsh environments. Intelligent flight control systems is an active area of research addressing limitations of PID control most recently through the use of reinforcement learning (RL), which has had success in other applications, such as robotics. Yet previous work has focused primarily on using RL at the mission-level controller. In this work, we investigate the performance and accuracy of the inner control loop providing attitude control when using intelligent flight control systems trained with state-of-the-art RL algorithms—Deep Deterministic Policy Gradient, Trust Region Policy Optimization, and Proximal Policy Optimization. To investigate these unknowns, we first developed an open source high-fidelity simulation environment to train a flight controller attitude control of a quadrotor through RL. We then used our environment to compare their performance to that of a PID controller to identify if using RL is appropriate in high-precision, time-critical flight control.

For decades, request-routing protocols operating at multiple layers of the network stack have been a staple of Internet services. Commonly deployed request-routing techniques use the requestor's IP address as an identifier of the client. For instance, using DNS as a request-routing protocol, the local DNS resolver's IP address is used as a surrogate identifier of the client in order to assign the client to the closest server. While such coarse associations may be acceptable for performance-centric purposes, they are not appropriate in settings that require fine-grained, enforceable bindings of clients to servers - e.g., to ensure that malicious clients are unable to bypass their bindings and issue their request to a server of their choosing. In this paper, we propose S3B (Software-defined Secure Server Bindings), a protocol that provides precise and enforceable client-server assignments. S3B uses a server module to assign clients unique access keys. Using HTTP redirection with the key encrypted as an additional domain label, the name server is able to distribute precise server assignments specific to each client. In addition, the server module maintains an access control list to enforce these assignments. As an implementation of the S3B protocol, we have developed an HTTP/S prototype and deployed it to Amazon AWS. Our performance evaluation suggests that our prototype introduces no discernible overhead for client requests. To evaluate S3B's effectiveness as a security appliance, we developed an application to isolate clients suspected as spiders, capable of virtually immediate containment once detected.

Similar to criminals in the physical world, cyber-criminals use a variety of illegal and immoral means to achieve monetary gains. Recently, malware known as ransomware started to leverage strong cryptographic primitives to hold victims' computer files "hostage" until a ransom is paid. Victims, with no way to defend themselves, are often advised to simply pay. Existing defenses against ransomware rely on ad-hoc mitigations that target the incorrect use of cryptography rather than generic live protection. To fill this gap in the defender's arsenal, we describe the approach, prototype implementation, and evaluation of a novel, automated, and most importantly proactive defense mechanism against ransomware. Our prototype, called PayBreak, effectively combats ransomware, and keeps victims' files safe.

Modern applications are often split into separate client and server tiers that communicate via message passing over the network. One well-understood threat to privacy for such applications is the leakage of sensitive user information either in transit or at the server. In response, an array of defensive techniques have been developed to identify or block unintended or malicious information leakage. However, prior work has primarily considered privacy leaks originating at the client directed at the server, while leakage in the reverse direction -- from the server to the client -- is comparatively under-studied. The question of whether and to what degree this leakage constitutes a threat remains an open question. We answer this question in the affirmative with Hush, a technique for semi-automatically identifying Server-based InFormation OvershariNg (SIFON) vulnerabilities in multi-tier applications. In particular, the technique detects SIFON vulnerabilities using a heuristic that overshared sensitive information from server-side APIs will not be displayed by the application's user interface. The technique first performs a scalable static program analysis to screen applications for potential vulnerabilities, and then attempts to confirm these candidates as true vulnerabilities with a partially-automated dynamic analysis. Our evaluation over a large corpus of Android applications demonstrates the effectiveness of the technique by discovering several previously-unknown SIFON vulnerabilities in eight applications.

The inability to identify the optimal construction of a nerve guidance conduit (NGC) for peripheral nerve regeneration is a challenge in the field of tissue engineering. This is attributed to the vast number of parameters that can be combined in varying quantities. A pre-existing normalization standard is applied in this paper which uses a calculated ratio of gap length divided by the graft's critical axon elongation denoted as L/Lc. This allows for a direct comparison of the nerve regenerative activity, a measure of performance, of any NGC across an array of gap lengths relative to a standard nerve conduit. Data was extracted from a total of 28 scientific publications that compared the nerve regenerative activity of experimental NGCs relative to standard NGCs. Of the extracted data, 40 parameters were identified that impacted the performance of the experimental conduits. We demonstrate how bootstrap aggregated neural networks provides substantial increases in accuracy in predicting the performance of a NGC over a single neural network and previous prediction attempts by the SWarm Intelligence based Reinforcement Learning (SWIRL) system. The improved accuracy will provide for a better understanding and insight for theorizing successful strategies for NGC development.