NESTiD Seminar
NESTiD Seminar
NESTiD Seminar Coordinator: George Mertzios
Here you can find the links to the dedicated webpages for the NESTiD Seminars and videos from previous years: Past NESTiD Seminars
NESTiD Seminar 2024/25
The seminar talks will be streamed online on zoom. Whenever the speaker is physically present in Durham, the presentation will also be in the room MCS2050 at the MCS building (in addition to zoom streaming). Whenever a speaker is not physically present, all people at Durham are invited to watch the zoom seminar in the above room. Below you can find the list of speakers and talks for the year 2024/25. In the schedule of talks below is in UK time.
NOTE: Please refer to the schedule below for any room changes for some selected talks.
Link for 2024-25 Seminar videos: click here
Term 1 of 2024/25:
(online on zoom)
Date and Time | Talk |
---|---|
Friday 11 Oct 2024 13:00 – 14:00 (Inaugural talk of the NESTiD seminar series) | Artur Czumaj (University of Warwick, UK) Streaming Graph Algorithms in the Massively Parallel Computation Model We initiate the study of graph algorithms in the streaming setting on massive distributed and parallel systems inspired by practical data processing systems. The objective is to design algorithms that can efficiently process evolving graphs via large batches of edge insertions and deletions using as little memory as possible. We focus on the nowadays canonical model for the study of theoretical algorithms for massive networks, the Massively Parallel Computation (MPC) model. We design MPC algorithms that efficiently process evolving graphs: in a constant number of rounds they can handle large batches of edge updates for problems such as connectivity, minimum spanning forest, and approximate matching while adhering to the most restrictive memory regime, in which the local memory per machine is strongly sublinear in the number of vertices and the total memory is sublinear in the graph sizes. This is a joint work with Gopinath Mishra and Anish Mukherjee. |
Friday 18 Oct 2024 13:00 – 14:00 | Jie Xu (University of Leeds, UK) A system for training massive-scale models with billions of parameters In this presentation, we will share our recent experience with designing and implementing a practical system, STRONGHOLD, for training massive-scale language models with billions of parameters, with a focus on our offloading mechanism for efficiently moving data amongst GPU memory and CPU RAM/secondary storages. Deep Learning is advancing rapidly, and with it, the size of foundation models is increasing exponentially. However, training these models requires significant GPU resources and power, which can be unaffordable for many academic and industry research teams. Even for AI teams in large companies, resources are limited, and purchasing and maintaining these devices can be prohibitively expensive. For instance, training a GPT-3 model requires over thousands of high-performance-configured A100 GPUs for continuous 3 months. Our system, STRONGHOLD, addresses this challenge by offloading model weights to CPU RAM or other secondary storages dynamically and loading them back when needed, minimizing GPU memory requirements. STRONGHOLD also allows data movement and on-GPU computation to overlap to hide the extra overhead introduced by the offloading mechanism. Compared to state-of-the-art offloading-based solutions, STRONGHOLD improves the trainable model size by 1.9x to 6.5x on a 32GB V100 GPU, with 1.2x to 3.7x improvement on the training throughput. We have successfully deployed STRONGHOLD in production to support large-scale DNN training. |
Tuesday 25 Oct 2024 13:00 – 14:00 | Arpan Mukhopadhyay (University of Warwick, UK) TBA TBA |
Friday 1 Nov 2024 13:00 – 14:00 | William (Billy) Moses Jr. (Durham University, UK) Towards communication-efficient peer-to-peer networks We focus on designing Peer-to-Peer (P2P) networks that enable efficient communication. Over the last two decades, there has been substantial algorithmic research on distributed protocols for building P2P networks with various desirable properties such as high expansion, low diameter, and robustness to a large number of deletions. A key underlying theme in all of these works is to distributively build a random graph topology that guarantees the above properties. Moreover, the random connectivity topology is widely deployed in many P2P systems today, including those that implement blockchains and cryptocurrencies. However, a major drawback of using a random graph topology for a P2P network is that the random topology does not respect the underlying (Internet) communication topology. This creates a large propagation delay, which is a major communication bottleneck in modern P2P networks. In this talk, we look at designing P2P networks that are communication-efficient (having small propagation delay) with provable guarantees. Our main contribution is an efficient, decentralized protocol, Close-Weaver, that transforms a random graph topology embedded in an underlying Euclidean space into a topology that also respects the underlying metric. We then present efficient point-to-point routing and broadcast protocols that achieve essentially optimal performance with respect to the underlying space. This talk is based on joint work with Khalid Hourani and Gopal Pandurangan that appeared as a paper in ESA 2024. |
Friday 8 Nov 2024 13:00 – 14:00 | Yannic Maus (TU Graz, Austria) TBA TBA |
Friday 15 Nov 2024 13:00 – 14:00 | Jason Schoeters (University of Cambridge, UK) TBA TBA |
Friday 22 Nov 2024 13:00 – 14:00 | TBA |
Friday 29 Nov 2024 13:00 – 14:00 | Shiri Chechik (Tel-Aviv University, Israel) TBA TBA |
Friday 6 Dec 2024 13:00 – 14:00 | Daniel Oi (University of Strathclyde, UK) TBA TBA |
Friday 13 Dec 2024 13:00 – 14:00 | TBA |
Term 2 of 2024/25:
(online on zoom)
Friday 17 Jan 2025 13:00 – 14:00 | Igor Razgon (Durham University, UK) Fixed parameter algorithms for computation of hypertreewidth parameters for restricted classes of hypergraphs TBA |
Friday 24 Jan 2025 13:00 – 14:00 | Cristina Chueca Del Cerro (Durham University, UK) Social networks’ structural property interdependencies: a (targeted) rewiring approach TBA |
Friday 31 Jan 2025 13:00 – 14:00 | Raouf Abozariba (Birmingham University, UK) Empirical wireless network measurement in indoor and outdoor environments TBA |
Friday 7 Feb 2025 13:00 – 14:00 | TBA |
Friday 14 Feb 2025 16:00 – 17:00 Please note the different time! | Sergio Rajsbaum (Universidad Nacional Autonoma de Mexico) TBA TBA |
Friday 21 Feb 2025 13:00 – 14:00 | Jia Hu (University of Exeter, UK) TBA TBA |
Friday 28 Feb 2025 13:00 – 14:00 | Hossein Anisi (University of Essex, UK) TBA TBA |
Friday 7 Mar 2025 13:00 – 14:00 | Oliver Kullmann (Swansea University, UK) TBA TBA |
Friday 14 Mar 2025 13:00 – 14:00 | TBA |
Friday 21 Mar 2025 13:00 – 14:00 | TBA |
Term 3 of 2024/25:
(online on zoom)
Friday 2 May 2025 13:00 – 14:00 | TBA |
Friday 9 May 2025 13:00 – 14:00 | TBA |