| SHARIF HOSSAIN FAKIR | 1561032: D, 中間発表 | ディペンダブルシステム学 | 井上 美智子,中島 康彦,大下 福仁,大和 勇太 |
|
title: A power Based Side-channel Analysis Technique for Hardware Trojan Detection in Integrated Circuit
abstract: Outsourcing of IC fabrication components has initiated the potential threat of design tempering using hardware Trojans and also has drawn the attention of government agencies and the semiconductor industry. The added functionality, known as hardware Trojan, poses major detection and isolation challenges. I will show two methods for hardware Trojan detection technique. First method magnifies the detection sensitivity for small Trojan in power-based side-channel analysis and requires golden fingerprints. A scan segmentation approach with a modified LOC test pattern application method is proposed. The proposed architecture allows activating any target segment of scan chain and keeping others freeze which reduces total circuit switching activity. The second method does not rely on golden fingerprints which is a self-referencing technique within same circuit. This method exhibits design for security (DFS) addressing clock tree based segmentation technique for scalability. After DFS, a self-authentication procedure is proposed in order to determine if a Trojan is inserted in a set of uninformed regions. The detection process uses parametric comparison of at least two neighboring regions, which consumes equal power for a set of selected patterns. We generate launch-on-capture test patterns and apply them with modification so as to restrict the switching activities (noises) from other regions. A theoretical analysis in the presence of die-to-die and intra-die process variations with the help of other existing methods is addressed. In our experiments, conducted for small Trojan circuits, we report a high detection rate thus substantiating its effectiveness in realizing an equal power self-authentication technique which is independent of any Golden IC. language of the presentation:English | |||
| 藤本 啓輔 | 1551092: M, 2回目発表 | コンピューティング・アーキテクチャ | 中島 康彦,井上 美智子,高前田 伸也,TRAN THI HONG |
|
title: Execution Throttling for Power-Constrained FPGA Accelerators
abstract: Power-constrained computing is now becoming essential paradigm in both high performance computing and embedded systems. Power budget is dynamically assigned to each computing resource for improving energy efficiency and system throughput. Modern computer systems have accelerator devices, such as GPUs and FPGAs, for higher energy efficiency and performance. Therefore power management mechanisms of such accelerator devices are required. In this paper, we present a lightweight mechanism of runtime power capping on FPGA systems. According to the amount of a given power budget, instead of the frequency scaling, the proposed mechanism controls the execution speed by throttling off-chip memory accesses from the computing logic, so that the power consumption is accordingly controlled. We evaluated the power controllability of the proposed mechanism by using an FPGA board with an embedded power meter. The result shows that the proposed approach has a high linearity of power control. The result also indicates that the accuracy of the power control depends on throttling interval granularities, and the control accuracy is improved by utilizing a longer throttling interval. Additionally, we compared the power control accuracy with a design-time fixed frequency scaling approach. The result shows that the proposed approach achieves the same accuracy as the static approach, even though the proposed approach allows the runtime power control. language of the presentation: Japanese | |||
| HUANG CHE | 1561026: D, 中間発表 | ソフトウェア設計学 | 飯田 元,藤川 和利,市川 昊平,渡場 康弘 |
|
title: An SDN-based Multipath Controller for Multiple TCP Streams
abstract: A large amount of scientific data needs to be transferred from one site to another as fast as possible in the computational science fields. High-speed data transfer between sites is very important, especially in the Grid computing field; GridFTP has been widely used for bulk data transfer over a wide area network. GridFTP achieves greater performance by supporting parallel TCP streams. Using parallel TCP streams improves the throughput of slow-start algorithms and lossy networks even on a single path. This research proposes a traffic engineering technique that increases the data transfer performance by using multiple paths simultaneously for the parallel TCP streams. For this purpose, we use Software-Defined Network (SDN) technology and its implementation, OpenFlow. This paper presents the design and implementation of the proposed system. Our performance evaluation demonstrates that our proposed system can accelerate GridFTP Transfer in both virtual and real global-scale environments. language of the presentation: Japanese | |||
| NAKASAN CHAWANAT | 1561029: D, 中間発表 | ソフトウェア設計学 | 飯田 元,藤川 和利,市川 昊平,渡場 康弘 |
|
title: Optimizing Multipath OpenFlow Controller for Distributed File Systems
abstract: Distributed file system (DFS) is one of the important components behind distributed systems. To store and process a large amount of data as well as maintain their integrity, many systems and features are implemented in DFSes to meet those demands. However, this evolution imposes additional demand on the network, which is usually the bottleneck in a DFS. By using multipathing techniques such as Multipath TCP, performance of the network could be improved. This work discusses how to optimize a multipath OpenFlow controller to further improve DFS performance. language of the presentation: English | |||
| U-CHUPALA PONGSAKORN | 1561034: D, 中間発表 | ソフトウェア設計学 | 飯田 元,藤川 和利,笠原 正治,市川 昊平,渡場 康弘 |
|
title: Increasing Datacenter Efficiency with Improved Task Scheduling and Communication
abstract: Nowadays, datacenter is incredibly important. Datacenter powers everything around us from using SNS to doing science with big data. Operating a datacenter is very expensive. Datacenter usage efficiency is vital to minimizing operational cost. With the simplify datacenter usage model, two factors that have major impact to datacenter efficiency are identified: task scheduling and communication. Container rebalancing method and application-aware network are developed to improve these two factors respectively. Application-Aware Network works by aligning applications' requirements to appropriate paths and routing each individual application accordingly. This network is implemented using OpenFlow, a de-facto standard implementation of Software-Defined Network. Container rebalancing is a scheduling method that load-balance actual utilization of each host in the datacenter in anticipation of future workload. This approach increase optimal overcommit ratio thus increasing datacenter utilization. language of the presentation: English | |||
| 岩根 史明 | 1451015: M, 2回目発表 | 数理情報学 | 池田 和司,佐藤 嘉伸,川人 光男,森本 淳 |
|
title: Relationship between passive movement speed and sensorimotor rhythm power recorded by dry-electrode EEG
abstract: Brain Robot Interface (BRI) has been researched and used all over the worlds. One of the main example of BRI use is BRI Rehabilitation. In BRI Rehabilitation, exo-skeleton robot moves impaired body parts of patients when there is specific features observed via electroencephalogram (EEG) signals. As you can imagine, robot moving time is limited in this conventional BRI Rehabilitation framework. In order to make it longer, we are planning to give continuous feedback during BRI Rehabilitation. The ultimate goal of our work is to provide proprioceptive feedback by modulating the wearable robot speed based on the EEG features recorded by a dry-electrode EEG system. In order to design a stable system, the first step is to investigate how different robot speed levels influence the sensorimotor rhythm power. To this end, we conducted an experiment with 4 subjects as of now. In the experiment, a 1-Degree-Of-Freedom robot worn at the level of the left elbow oscillated at different speed levels at each trial, while the subject’s brain activity was recorded. In order to have a point of reference, additional trials were collected where subjects rested, or performed motor execution of their left arm at 1 Hz. As a result, we found that there was no change of EEG modulation even when robot speed changed. Moreover, power of both alpha and beta band during passive movement were stronger than that of motor imagery.
language of the presentation: English | |||
| SORIANO JAYMAR BANGAYAN | 1561033: D, 中間発表 | 数理情報学 | 池田 和司,佐藤 嘉伸,久保 孝富 |
|
Title: Focal Brain Cooling in Epilepsy: Insights From a Temperature-Dependent Neural Mass Model Abstract: Focal brain cooling has been extensively performed with experiments in animal and some human subjects and has been shown to suppress epileptic activity. While this therapeutic treatment can be done viably using an implantable cooling device, a precise mechanism of how cooling suppresses the epileptic activity is not yet clearly understood alongside observation that neurotransmitter concentrations are reduced during cooling. In light of this, we formulated temperature-dependence in a neural mass model to capture the effect of cooling in suppressing epileptic discharges in rats. The effect of reduced neurotransmitter concentration was modelled using a temperature coefficient to attenuate post-synaptic impulse response function. This however does not completely capture experimental observations. We propose that a homeostatic mechanism plays a role to compensate reduced average firing as a result of reduced average post-synaptic potential. This mechanism involves lowering the average threshold potential of firing and increasing its variability across neuronal population.The proposed model is able to capture suppression of epileptic discharges with parameters estimated from cooling experiments with rats. Language of the presentation: English | |||
| 諏訪部 開 | 1551056: M, 2回目発表 | 数理情報学 | 池田 和司,佐藤 嘉伸,川人 光男,森本 淳 |
|
title: Neural decoding of thought contents during mind wandering abstract: We spend half of our waking time with thoughts independent of what is going on here and now, which is known as mind wandering. While contents of such self-generated task-unrelated thoughts are known to be associated with past experiences and future plans, how neural representations of those thought contents are different and common to those of task-related thoughts still remains unknown. Here, to reveal the commonalities of neural representations between task-related and task-unrelated thoughts, we performed brain decoding of thought contents (object or scene categories) during mind wandering by applying decoders trained on stimulus-induced and imagery-induced brain activity patterns (stimulus- and imagery-trained decoders, respectively). We conducted 28 sessions of thought sampling experiments while measuring brain activity by functional magnetic resonance imaging, and collected 138 reports of task-unrelated thought. Our analyses demonstrated that the contents of mind wandering were predicted with moderate accuracy using both of the stimulus-trained and imagery-trained decoders trained on brain activities in higher visual cortex and several areas associated with default mode network. While the results are preliminary, they may suggest common neural representations between task-related and task-unrelated thoughts, and open the possibility to capture richer information associated with self-generated task-unrelated thoughts based on neural similarities with task-related thoughts. language of the presentation: Japanese | |||
| 本田 新 | 1551098: M, 2回目発表 | 数理情報学 | 池田 和司,杉本 謙二,佐々木 博昭 |
|
title: Statistical mechanics of Perceptron learning with noisy teacher
abstract:Statistical mechanics method can apply to information problem such as the image restoration and neural network, and that the framework is called information statistical mechanics. Information statistical mechanics can analyze the perceptron learning algorithm. The perceptron learning algorithm is an on-line learning algorithm that the perceptron updates its connection weight to approach hopeful output (teacher's signal). Miyoshi extended the analysis to the ensemble learning and/or noisy cases. Among them, the learning curve(the learning process) has specific behavior when the teacher's signal has observation noise. In this study we derive the learning curve of perceptron with noisy teacher's signal using statistical mechanics, and analyze the behavior around the convergence point to explain why and how occurs it. In addition, we analyze the AdaTron learning and the Hebbian learning which include the on-line learning. language of the presentation: Japanese | |||