A Gentle Introduction to Hardware Performance Counters

Your CPU, graphics card, even memory bus is equipped with special registers, that monitor events in your execution units. Using these counters, you can conduct non-invasive, high-resolution measurements of e.g. the number of branches, cache-accesses, loading stalls, and special issued instructions. This talk gives you a hands-on overview of existing Hardware Performance Counters on recent Intel and ARM platforms and how to read them using tools like Linux perf, or libpapi. I will show you common pitfalls when evaluating your results and teach you how to annotate your program to find the most expensive bottlenecks on an instruction level.

Introductory Talk – Research Background (Business Process Modeling and Execution)

Enterprise Resource Planning (ERP) is a business process management system in which integrated applications are used to manage business processes in a shared data environment. Business Process Modeling Notation (BPMN) is frequently utilized to simplify the implementation complexity of ERP applications. This talk gives you a hands-on overview of my research work in this domain. I will show our proposed framework that automatically generates executable web services for ERP exchange and conversion processes from BPMN Models. As a part of research, BPMN to Web services transformation (B2W) tool is developed.

Transcalibur: A Weight Shifting Virtual Reality Controller for 2D Shape Rendering

We propose Transcalibur, a hand-held VR controller that can render a 2D shape by changing its mass properties on a 2D planar area. We built a computational perception model using a data-driven approach from the collected data pairs of mass properties and perceived shapes. This enables Transcalibur to easily and effectively provide convincing shape perception based on illusory effects. Our user study showed that the system succeeded in providing the perception of various desired shapes in a virtual environment. For more detail of this project please see: https://jotaros.github.io/project/transcalibur/

Multi-Tenancy Concepts for FPGA Accelerators

To optimize the throughput of FPGA accelerator resources and facilitate their applicability in realistic applications featuring multiple concurrent tasks, multi-tenancy is a desirable feature for our research prototype Metal FS. Based on the current single-tenant execution model, we will adapt the operator concept and data stream infrastructure to handle data flows from multiple different task contexts at the same time. Approaches for multi-tenancy must provide mechanisms to avoid or control the potential of isolation breaches between independent contexts. Furthermore the performance impact of processing multiple concurrent tasks on a single operator implementation is an open question, probably with varying answers depending on the task characteristics or operator architecture. We expect to gain valuable insights regarding the feasibility of multi-tenancy on FPGA accelerators, i.e. whether the platform can sustain the additional task-management overhead, and also regarding which operator characteristics lend themselves well to different task-switching granularities. In this presentation I will discuss multi-tenancy approaches for Metal FS and their associated costs and design tradeoffs.

Discovering Data Models From Event Logs

Business process mining is becoming an increasingly important field for understanding the behavioral perspective of any given organization. In a process mining project, process experts are tasked with discovering or improving the operational business processes. They do soby analyzing event logs, the starting point of any process mining endeavor. Despite event logs capturing behavioral information, we argue thatthey are also a rich source of domain specific information. This informationis not represented explicitly in a process model but, nevertheless,it provides valuable contextual information. To this end, we propose anapproach to discover a data model that complements traditional process mining techniques with domain specific information. The approach isevaluated in terms of feasibility by being applied to two real-life event logs.

Digital Twins for Indoor Built Environments

One of the key challenges in modern Facility Management (FM) is to digitally reflect the current state of the built environment, referred to as-is or as-built versus as-designed representation. While the use of Building Information Modeling (BIM) can address the issue of digital representation, the generation and maintenance of BIM data requires considerable amount of manual work and domain expertise. Another key challenge is being able to monitor the current state of the built environment, which is used to provide feedback and enhance decision making. The need for an integrated solution for all data associated with the operational lifecycle of a building is becoming more pronounced as practices from Industry 4.0 are currently being evaluated and adopted for FM use. This research presents an approach for digital representation of indoor environments in their current state within the lifecycle of a given building. Such an approach requires the fusion of various sources of digital data. The key to solving such a complex issue of digital data integration, processing and representation is with the use of a Digital Twin (DT). A DT is a digital duplicate of the physical environment, states, and processes. A DT fuses as-designed and as-is physical representations, typically in the form of floorplans, point cloud and BIM data, with additional information layers pertaining to the current and predicted states of an indoor environment or complete building (e.g., sensor data). The design, implementation and initial testing of a prototypical DT software platform for indoor environments is presented and described. The DT software platform is implemented using a Service Oriented Architecture (SOA) paradigm, and its feasibility is presented through functioning and tested key software components within prototypical SOS implementations. The main outcome of this research shows that key data related to the built environment can be semantically enriched and combined to enable digital representations of indoor environments, based on the concept of a DT. Furthermore, the outcomes of this research show that digital data, related to FM and Architecture, Construction, Engineering, Owner and Occupant (AECOO) activity, can be combined, analyzed and visualized in real-time using a service-oriented approach. This has great potential to benefit decision making related to Operation and Maintenance (O&M) procedures within the scope of the post-construction lifecycle stages of typical office buildings.

infill: Automatic Reinforcement of Laser-Cut Closed Box Structures

We present infill, a software tool that identifies and resolves points of potential failure in laser cut closed box structures by automatically reinforcing them. Earlier work suggested that load bearing objects such as tables and chairs can be made using these structures, however, we found many objects created in this vein to break under minimal load. The reason is that finger joints are weak against tension loads and any non-convex object can allow users to apply tension at these joints. Infill uses a graph-based algorithm to classify potential points of failure in an object and reinforces them by extending and anchoring plates into the internal space of the model, ensuring that finger joints are only subjected to shearing, which they are strong against. Infill also ensures that the resulting structure can still be assembled physically. Infill is designed to use little computational resources, allowing it to run in the background during editing, reinforcing the user’s model continuously and automatically. We integrated infill into the Kyub editor for laser cutting. Our survey indicates that 71% (284/400) of user generated models in its repository would benefit from reinforcements generated by infill. In our technical evaluation, we found that reinforced objects took between 3 to 52 times more force to break than without.

EvaP – Course Evaluation at HPI

Course evaluation is an established process at many universities. Responsible teachers and tutors receive feedback about their courses and might get ideas how the knowledge transfer could be improved. Additionally, personal evaluation results are often required for academic applications.

Traditionally, course evaluation was done with paper questionnaires which students completed during one of the last lectures of a term. Nowadays, online platforms enable giving digital feedback, resulting in faster data processing and less manual work.

However, not all institutions can impress with well-designed processes and often lack a relevant participation count in their evaluations. This talk will highlight why course evaluations at HPI outperform those of most other institutions and will not only give an insight into UX and management processes but also into technical details of the open-source project EvaP and the continuously successful active development of this evaluation tool. You will also learn how you can benefit most of evaluations of your own courses.

Space Independent Real Walking in Virtual Reality

The goal of my research is to allow virtual reality experiences to be run in arbitrary tracking volumes and with arbitrary physical objects. VR experiences today are designed with a specific tracking volume and objects in mind, such as “square 5x5m space with a rubber sword”. This prevents experiences from running with different objects or in tracking volumes of smaller size or different shape, making it impossible to share experiences, especially with home users. I address this by creating an abstraction between VR applications and the space and physical objects they are using. Instead of accessing space and physical objects directly, in my system applications express their needs in an abstract way, which my systems then maps to the actual available physical space and physical objects. This allows VR applications to run on a wide range of installations. Solving this problem would have substantial commercial impact, as the proliferation of real-walking VR is currently hindered by developers’ reluctance to require users to have space and objects. My work is inspired by operating systems research. Before opening systems, application programs were written for a specific machine. Operating systems allow applications to run on arbitrary computers and architectures by creating an abstraction of the physical hardware, an API, that allows applications from accessing the hardware directly.

Towards Joint Design-Time and Run-time Verification of Complex system

In this talk, I will describe a new approach that combines the backward design time and forward run-time checking techniques. The backward checking is used to find out all counterexample paths that will lead to an unsafe state and establish an unsafe area with a radius of k. At run-time, the forward model checking will periodically perform n steps checking from the current state. Associated with the verification result from the development phase, systems can perceive potential hazards k+n-1 steps before their happening. This approach moves the most computationally intensive task to the design time and obtains enough confidence and reactive time using constrained time and resources. For systems that cannot acquire enough information to establish precise model during development, our work provides a complete verification method that can help improve the system, generate the unsafe areas towards unsafe states, and cope with the failure at run-time.

Constructing geo-referenced virtual city models from point cloud primitives

In this talk I will introduce a novel approach to constructing spatially referenced, multidimensional virtual city models from generated point clouds for areas for which no reliable geographic reference data is available. A multidimensional point cloud is an unstructured arrangement of single, irregular points in a spatial 3D coordinate system with time stamp. If georeferenced, due to their characteristics point clouds represent a highly precise reference data set and can be used as a basic data set in a variety of applications, especially for the representation of surfaces, structures, terrain and objects. Point clouds are used here to create a virtual 3D city model that represents the complex, granular cityscape of Jerusalem and its center, the Old City. The gathered spatio-temporal data require further processing in order to extract geographical features and semantic characteristics at a certain resolution and scale and thus allow for the meaningful utilisation in applications. In this talk, an insight into the processing of an unstructured point cloud is given in order to extract and classify the 3D city structure and reconstruct its objects.

Intrinsic Decomposition of RGB-D Images on Smartphones

In this talk, I will present my current research topic, "Intrinsic Decomposition of RGB-D Images on Smartphones": Intrinsic decomposition refers to the problem of estimating scene characteristics such as albedo and shading provided one view or multiple views of a scene. The inverse problem setting, where multiple unknowns are solved given a single known pixel-value, is highly under-constrained. When provided with correlating image and depth data, intrinsic scene decomposition can be facilitated using depth-based priors, which nowadays is easy to acquire with high-end smartphones by utilizing their depth sensors. Unlike most of the previous methods that assume only diffuse reflectance, we consider both diffuse and specular pixels. Our GPU-based visual processing implemented on an iPhone via Metal API allows for interactive performance. The above decomposition allows for a variety of applications such as recoloring, relighting, appearance editing, and stylization.

Extended Bayesian and Contextual Multi-Armed Bandit Models for Sequencing Code Inspection Tasks

Code inspection tasks consist of reading the source code in search for problems, for instance, coding rule violations and potential software bugs. These inspections tasks are central to various time consuming activities like code reviews and software debugging. Software debugging, in particular, was shown by different studies to consume from 20% to 50% of programmers' time. I investigated how to minimize this cost by sequencing code inspection tasks in a way that a crowd of programmers could quickly locate all known bugs in given popular open source projects. My approach was to model this problem as a sequential decision process that balances exploration and exploitation of source code statements with respect of their probability of presenting a bug. I implemented this sequential decision process as different instances of Bayesian and contextual multi-armed bandits models. I also extended these models to take into account prior knowledge of the causal relationships that I uncovered between task execution factors (e.g., programmer confidence, code complexity) and the accuracy of these tasks. My results are promising in a sense that the tasks sequenced by the improved bandit models located all bugs with only 20% of the available code inspection tasks.