• Leverage design space exploration tool to create an efficient accelerator for DNNs. • Reuse redundant hardware components to minimize chip area. • Schedule to minimize dataflow transitions to maximize efficiency. This paper proposes a new dataflow-flexible accelerator design that addresses the limitations of existing heterogeneous dataflow accelerator (HDA) for handling the computation of multiple deep neural network (DNN) models. The design offers increased dataflow flexibility and higher efficiency compared to existing works. The accelerator utilizes a fixed set of representative dataflows implemented as operating modes and switches between them dynamically. A design space exploration (DSE) tool is leveraged to evaluate the efficiency of candidate dataflows and determine the optimal number and types of operating modes. Each layer of the target DNN models is assessed with different operating modes to select the optimal mode for each layer. Also, two supplementary optimization techniques are adopted to reduce the overheads from supporting a multitude of dataflows. One optimizes to minimize the number of transitions of dataflows, which incur severe overheads. The other optimizes to maximize the reuse of hardware components associated with supporting multiple dataflows. By identifying the redundant hardware components, the proposed design minimizes the chip area, another aspect where dataflow-flexible accelerators suffer. Experimental results demonstrate that our algorithm achieves greater dataflow flexibility with high efficiency, Compared to HDA, our design is, on average, 34.6 % lower in latency at the cost of 6.4 % area and negligible energy overhead.

Most home appliance devices such as electric ovens are reactive systems which repeat receiving a user input/event through an event handler, updating their internal state based on the input, and generating outputs. A challenge to test a reactive program is to check if the program correctly reacts to various non-deterministic sequence of events because an unexpected sequence of events may make the system fail due to the race conditions between the main loop and asynchronous event handlers. Thus, it is important to systematically generate/test various sequences of events by controlling the order of events and relative timing of event occurrences with respect to the main loop execution. In this paper, we report our industrial experience to solve the aforementioned problem by developing a systematic event generation framework based on concolic testing technique. We have applied the framework to a LG electric oven and detected several critical bugs including one that makes the oven ignore user inputs due to the illegal state transition.

In this paper, we propose an effective end-to-end virtual path construction system, which exploits path diversity over heterogeneous wireless networks. The goal of the proposed system is to provide a high quality live video streaming service over heterogeneous wireless networks. First, we propose a packetization-aware fountain code to integrate multiple physical paths efficiently and increase the fountain decoding probability over wireless packet switching networks. Second, we present a simple but effective physical path selection algorithm to maximize the effective video encoding rate while satisfying delay and fountain decoding failure rate constraints. The proposed system is fully implemented in software and examined over real WLAN and HSDPA networks.

In this paper, we propose a Luby transform encoding pattern-aware symbol packetization algorithm to minimize the quality degradation of video streaming service caused by packet losses over wireless network. To achieve this goal, the relationship among Luby transform encoded symbols is analyzed based on Luby transform encoding pattern, and the proposed packetization algorithm is designed to minimize packet loss effects by reducing the dependency among packets conveying Luby transform encoded symbols. Finally, experimental results are provided to show the performance of the proposed algorithm.