Friday, Apr. 21st, 2023 at 16:45-18:15
The URL of Zoom will be announced via the faculty’s mailing list 3 days before the colloquium.
The lecture material will be available in PandA at 16:15 on the day of the colloquium.
Dr. Takuya YASUI
Innovation by Design and Technology Co-Optimization
Although the technology scaling of the logic CMOS process tends to slow down, it is still in progress even now in the single-digit node era. It has been realized that the performance-power-area (PPA) is continuously migrated by Design Technology Co-Optimization (DTCO). In particular, Standard Cell are extremely important to achieve the full scale node scaling from N5 to N3. In this talk, FinFlex Technology which TSMC developed new standard cell architecture will be introduced.
2023年4月21日(金) 16:45-18:15
ZoomのURLは開催3日前に研究科メーリングリストでお知らせします．
講演資料は開催当日の16時15分にPandAに掲載します．
安井 卓也 様
設計とテクノロジの協調最適化によるイノベーション
ロジックCMOSプロセスのスケーリング(微細化)は鈍化の傾向にあるものの、1桁ノード時代となった現在もなお進行中である。これまで微細化と共に様々な課題に直面してきたが、設計とテクノロジの協調最適化Design Technology Co-Optimization（DTCO）により,面積縮小とその性能向上を実現してきた。特に先端スタンダードセルは5nmから3nmでのフルスケールノードのスケーリングを実現するためにより重要になってきている。本講演では、TSMCが開発した新スタンダードセル構造 FinFLEX technologyの開発事例を紹介する。
Friday, May 19th, 2023 at 16:45-18:15
The URL of Zoom will be announced via the faculty’s mailing list 3 days before the colloquium.
The lecture material will be available in PandA at 16:15 on the day of the colloquium.
Prof. Tasuku HIRAISHI
A Parallel Language with Backtracking-based Load Balancing for Irregular Applications
When we implement parallel programs for high-performance computing environments such as computing clusters or supercomputers, the issue of load imbalance is often encountered, especially for irregular applications, which are difficult to divide into subproblems with approximately equivalent loads. Task-parallel languages are often used to implement such applications, which enable dynamic spawning of tasks to be automatically assigned to workers. We are currently developing a task-parallel language called Tascell that employs a unique dynamic load balancing mechanism called backtracking-based load balancing. In Tascell, a worker initially chooses not to spawn a task and instead performs sequential computations. When a worker receives a task request, it spawns a task as if it had made a different choice in the past. In this presentation, I will introduce the design and implementation of Tascell, as well as its applications, such as graph mining and hierarchical matrix (H-matrix) calculations.
2023年5月19日(金) 16:45-18:15
ZoomのURLは開催3日前に研究科メーリングリストでお知らせします．
講演資料は開催当日の16時15分にPandAに掲載します．
平石 拓 先生
不規則アプリケーション向けのバックトラックに基づく負荷分散機能を備えた並列言語
計算機クラスタやスーパーコンピュータなどの高性能計算環境において並列プログラムを実装する際，負荷の不均衡の問題に遭遇することがある．この問題は不規則アプリケーション，すなわち負荷がほぼ等しい部分問題に分割することが困難なアプリケーションにおいて特に問題となる．このようなアプリケーションを効率的に実装するため，タスクを動的に生成しておくとそれらがワーカに自動的に割り当てられるという機能を備えたタスク並列言語が用いられることがある．我々は現在、バックトラック型負荷分散と呼ぶ独自の動的負荷分散機構を持つタスク並列言語Tascellを開発している．Tascellでは，ワーカーは最初はタスクを生成せず逐次計算を実行するが，タスク要求を受け取ると、あたかも過去に別の選択をしていたかのようにタスクを生成する．本発表では，Tascellの設計と実装，およびグラフマイニングや階層型行列（H行列）計算などの応用例を紹介する。
Friday, Jun. 16th, 2023 at 16:45-18:15
The URL of Zoom will be announced via the faculty’s mailing list 3 days before the colloquium.
The lecture material will be available in PandA at 16:15 on the day of the colloquium.
Akihiro Suda
The internals and the latest trends of container runtimes
Containers are a set of various lightweight methods to isolate filesystems, CPU resources, memory resources, system permissions, etc. Containers are similar to virtual machines in many senses, but they are more efficient and often less secure. This talk roughly consists of the following three parts:
1. Introduction to containers and how they spread in the last decade
2. Internals of container runtimes: namespaces, cgroups, capabilities, seccomp, etc.
3. Latest trends: Non-Docker containers, User Namespaces, Rootless Containers, Kata Containers, gVisor, WebAssembly, etc.
2023年6月16日(金) 16:45-18:15
ZoomのURLは開催3日前に研究科メーリングリストでお知らせします．
講演資料は開催当日の16時15分にPandAに掲載します．
須田 瑛大 様
コンテナランタイムの仕組み及び最新の動向
コンテナとは、ファイルシステム、CPU資源、メモリ資源、システム管理権限等を分離するための種々の軽量な手法を寄せ集めたものである。コンテナは仮想マシンに似ている面も多いが、効率面で優る。一方、セキュリティ面では劣ることもある。本講演は、主に次の3部で構成する:
1. コンテナの紹介および、コンテナが近年に至って普及した経緯
2. コンテナランライムの仕組み: namespaces、cgroups、capabilities、seccomp 等
3. 最新の動向: Docker以外のコンテナ、User Namespaces, Rootless Containers, Kata Containers, gVisor, WebAssembly 等
Friday, July 21st, 2023 at 16:45-18:15
The URL of Zoom will be announced via the faculty’s mailing list 3 days before the colloquium.
The lecture material will be available in PandA at 16:15 on the day of the colloquium.
Prof. Mirai Ikebuchi
Assistant professor at Dept. of Communications and Computer Engineering
Mirai Ikebuchi received a PhD degree in computer science from Massachusetts Institute of Technology in 2021. Then she moved to National Institute of Informatics in Japan as a postdoctoral researcher, and obtained the current position on April 2023. She works on programming languages, software verification, and mathematics around them.
Introduction to Proving the Correctness of Programs
There are several ways to make sure that a program is correct. One of the most commonly employed way is testing. Even though testing sounds handy, making good test cases is in fact quite difficult, and bad tests can overlook bugs in the program. As an approach to completely ensure a certain property of a program, I introduce Coq, a tool in which we can write formal proofs and the validity of proofs is automatically checked. In the talk, I will begin with a short introduction on how to write programs and proofs in Coq, and then give some examples appearing in practical situations.
2023年7月21日(金) 16:45-18:15
ZoomのURLは開催3日前に研究科メーリングリストでお知らせします．
講演資料は開催当日の16時15分にPandAに掲載します．
池渕未来 先生
通信情報システム専攻コンピュータ工学講座 助教
プログラムの正しさについての証明入門
プログラムが正しいことを確認する方法はいくつかある。最も一般的に採用されている方法のひとつとしてテストがある。 テストというと便利そうに聞こえるが、良いテストケースを作るのは実際にはかなり難しい。 プログラムのある性質を完全に保証するためのアプローチとして、Coqを紹介する。Coqは、形式証明を書くことができ、証明の妥当性が自動的にチェックされるツールである。 本講演では、Coqを用いたプログラムと証明の書き方について簡単に紹介した後、実用的な場面での例や、実際にCoqが活用されている事例をいくつか紹介する．
Friday, Oct. 20th, 2023 at 16:45-18:15
The URL of Zoom will be announced via the faculty’s mailing list 3 days before the colloquium.
The lecture material will be available in PandA at 16:15 on the day of the colloquium.
Dr. Akihito Suzuki
Network Resource Management Using Multi-Agent Deep Reinforcement Learning
The future evolution of networks will require various emerging communication technologies, such as Network Functions Virtualization (NFV) and Edge Computing (EC). NFV enables telecommunication providers to offer multiple network services by combining virtual network functions (VNFs). EC enhances computing capabilities at the network edges, enabling efficient task offloading. However, both technologies face challenges in optimizing resource allocation within the constraints of limited network and server resources. This presentation introduces network resource management methods using multi-agent deep reinforcement learning (MADRL) to rapidly find near-optimal resource allocations in future networks incorporating NFV and EC.
2023年10月20日(金) 16:45-18:15
ZoomのURLは開催3日前に研究科メーリングリストでお知らせします．
講演資料は開催当日の16時15分にPandAに掲載します．
鈴木 晃人 様
マルチエージェント深層強化学習によるネットワークリソース管理
将来ネットワークの進化には，ネットワーク機能仮想化（NFV）やエッジコンピューティング（EC）など，新たな通信技術が必要になる．NFVは，仮想ネットワーク機能（VNF）を組み合わせることで，様々なネットワークサービスの提供を可能にする．ECは，ネットワーク・エッジのコンピューティング能力を強化し，効率的なタスク・オフロードを可能にする．両技術では，ネットワークやサーバーのリソースが限られる中で，最適なリソース割り当てを見つけることが重要な研究課題となっている．本発表では，NFVやECを想定した将来ネットワークにおいて，効率的なリソース割り当てを迅速に見つけるためのマルチエージェント深層強化学習を用いたネットワークリソース管理手法を紹介する．
Friday, Nov. 17th, 2023 at 16:45-18:15
The URL of Zoom will be announced via faculty’s mailing list 3 days before the colloquium.
The lecture material will be available in PandA at 16:15 on the day of the colloquium.
Prof. Youngwoo Kim
Post Moore’s Law Era: Semiconductor Packaging Technology based System Scaling
Recently, transistor scaling represetned by Moore’s law is facing the limit. As a superior alternative, semiconductor packaing technology based system scaling has been proposed. 2.5/3 D integration based on packaging technology provies various benefis such as low-power consumption, small form-factor, and high system bandwidth. However, advanced packaing technology has also faced various limits such as extremely high design complexity, high-frequency noises, computational resource, and etc. In this talk, the definition of the packaing technology will be introduced with industrial difficulties. Also, various reaerches to solve the limit will be introdced considering signal/power integrity.
2023年11月17日(金) 16:45-18:15
ZoomのURLは開催3日前に研究科メーリングリストでお知らせします．
講演資料は開催当日の16時15分にPandAに掲載します．
Prof. Youngwoo Kim
Post Moore’s Law Era: Semiconductor Packaging Technology based System Scaling
Recently, transistor scaling represetned by Moore’s law is facing the limit. As a superior alternative, semiconductor packaing technology based system scaling has been proposed. 2.5/3 D integration based on packaging technology provies various benefis such as low-power consumption, small form-factor, and high system bandwidth. However, advanced packaing technology has also faced various limits such as extremely high design complexity, high-frequency noises, computational resource, and etc. In this talk, the definition of the packaing technology will be introduced with industrial difficulties. Also, various reaerches to solve the limit will be introdced considering signal/power integrity.
Friday, Dec. 15th, 2023 at 16:45-18:15
The URL of Zoom will be announced via the faculty’s mailing list 3 days before the colloquium.
The lecture material will be available in PandA at 16:15 on the day of the colloquium.
Prof. Yuto Nakashima
Dictionary compressors and repetitiveness measures
In lossless text compression, dictionary compressors are effective for highly repetitive texts, and many dictionary compressors have been proposed. For example, LZ77, the core of gzip, and grammar compression are well-known. However, a good model for capturing the string repetitiveness has yet to be known. Motivated by this fact, there is much research to theoretically understand string repetitiveness by discussing the relations between compression sizes (which we call repetitiveness measures) for each compressor. In this talk, I will introduce various dictionary compressors and recent research about repetitiveness measures.
2023年12月15日(金) 16:45-18:15
ZoomのURLは開催3日前に研究科メーリングリストでお知らせします．
講演資料は開催当日の16時15分にPandAに掲載します．
中島 祐人 先生
文字列の辞書式圧縮と反復性指標
文字列の可逆圧縮において，繰り返し（反復）を多く含む文字列の圧縮には，文字列中の繰り返し情報に着目した「辞書式圧縮」と呼ばれる圧縮の枠組みが効果的であり，多くの辞書式圧縮法が提案されている．例えば，gzipなどの核となるLZ77や，文法圧縮などがその代表として挙げられる．しかし，文字列中の繰り返しを捉える良いモデルは明らかになっていない．そこで，各辞書式圧縮法による圧縮サイズ（反復性指標）の関係を明らかにすることで，反復性を理論的に理解することを目的とした研究が盛んである．本講演では，様々な辞書式圧縮法や，反復性指標に関する最近の研究動向について紹介する．
Friday, Jan. 19th, 2024 at 16:45-18:15
The URL of Zoom will be announced via the faculty’s mailing list 3 days before the colloquium.
The lecture material will be available in PandA at 16:15 on the day of the colloquium.
Prof. Yu-Guang Chen
Navigating Aging Effects: Concepts and Implementation in Reliable Computing Systems
As CMOS technology undergoes further scaling down, the aging effect emerges as a non-negligible threat to the lifetime reliability of computing systems, potentially causing performance degradation or timing failures. Addressing these challenges requires a thorough understanding of how aging effects influence the results of modern computing systems, leading to the development of appropriate methodologies for aging detection, mitigation, and tolerance. In this presentation, I will provide a concise overview of major aging effects and their causes. I will then delve into two key aspects: (1) Aging-aware, energy-efficient task deployment for heterogeneous multicore systems, and (2) Aging-aware SRAM-based Computing-In-Memory architecture designed for multiply accumulate operations. Throughout, I will showcase innovative concepts proposed by our research team to tackle these challenges and discuss the implementation difficulties we’ve encountered. The aim is to furnish the audience with a foundational background in reliable computing system design and to inspire more researchers to contribute to this evolving field.
2024年1月19日(金) 16:45-18:15
ZoomのURLは開催3日前に研究科メーリングリストでお知らせします．
講演資料は開催当日の16時15分にPandAに掲載します．
Prof. Yu-Guang Chen
Navigating Aging Effects: Concepts and Implementation in Reliable Computing Systems
As CMOS technology undergoes further scaling down, the aging effect emerges as a non-negligible threat to the lifetime reliability of computing systems, potentially causing performance degradation or timing failures. Addressing these challenges requires a thorough understanding of how aging effects influence the results of modern computing systems, leading to the development of appropriate methodologies for aging detection, mitigation, and tolerance. In this presentation, I will provide a concise overview of major aging effects and their causes. I will then delve into two key aspects: (1) Aging-aware, energy-efficient task deployment for heterogeneous multicore systems, and (2) Aging-aware SRAM-based Computing-In-Memory architecture designed for multiply accumulate operations. Throughout, I will showcase innovative concepts proposed by our research team to tackle these challenges and discuss the implementation difficulties we’ve encountered. The aim is to furnish the audience with a foundational background in reliable computing system design and to inspire more researchers to contribute to this evolving field.