Fall 2022 Schedule

Mon, Aug 29	GHC 6501		available
Fri, Sep 02	GHC 6501		available
Mon, Sep 05	GHC 6501		available
Fri, Sep 09	GHC 6501		available
Mon, Sep 12	GHC 6501		available
Fri, Sep 16	GHC 6501		available
Mon, Sep 19	GHC 6501		available
Fri, Sep 23	GHC 6501		available
Mon, Sep 26	GHC 6501		available
Fri, Sep 30	GHC 6501		available
Mon, Oct 03	GHC 6501		available
Fri, Oct 07	GHC 6501		available
Mon, Oct 10	GHC 6501		booked
Fri, Oct 14	GHC 6501		available
Mon, Oct 17	GHC 6501		available
Fri, Oct 21	GHC 6501		Not available
Mon, Oct 24	GHC 6501		available
Wed, Oct 26	GHC 6501		available
Fri, Oct 28	GHC 6501		Detection and Location Uncertainty in Multi-agent Active Search Presented by Arundhati Banerjee https://cmu.zoom.us/j/91744959037?pwd=Nm85VzE3S0ZHWjVuRHo5NjYxeXVNQT09 Active search, in applications like environment monitoring or disaster response missions, involves autonomous agents detecting targets in a search space using decision making algorithms that adapt to the history of their observations. Active search algorithms must contend with two types of uncertainty: detection uncertainty and location uncertainty. The more common approach in robotics is to focus on location uncertainty and remove detection uncertainty by thresholding the detection probability to zero or one. In contrast, it is common in the sparse signal processing literature to assume the target location is accurate and instead focus on the uncertainty of its detection. In this talk, I will present a multi-agent active search algorithm that can jointly handle both target detection and location uncertainty for state estimation from noisy sensor measurements, without the need for a central controller or synchronous inter-agent communication. Our results indicate noticeable improvements in target recovery performance over baselines that only account for either target detection or location uncertainty separately. I will also demonstrate the real world transferability of our algorithm using a realistic simulation environment created on the Unreal Engine platform.
Mon, Oct 31	GHC 6501		Memento: Architectural Support for Ephemeral Memory Management Presented by Ziqi Wang Serverless computing is an increasingly attractive cloud paradigm due to its ease of use and fine-grained pay for-what-you-use billing. However, serverless computing poses new challenges to system design due to its short-lived execution model. Our detailed analysis reveals that memory allocations in serverless functions are typically small and ephemeral: i.e., freed shortly after allocation. Unfortunately, these functions pay the full critical-path costs of memory management in both user space and the operating system without the opportunity to amortize these costs over their short lifetimes. In this talk, I will present Memento, a hardware-centric design to management that alleviates the overheads of ephemeral memory management. Memento achieves its design goal with two key mechanisms. The first is a hardware object allocator that performs in-cache object allocation and free operations. The second is a hardware page allocator that manages a small pool of physical pages and handles physical memory allocation. We evaluate Memento with full-system simulations across a diverse set of serverless workloads. The results show that, on average, achieves a 14% (and up to 22%) speedup of function execution time, and reduces the runtime pricing cost by 24%. Memento also eliminates most of the memory management work from the critical path and almost accomplishes the theoretically optimal result that any optimization scheme can achieve.
Fri, Nov 04	GHC 6501		available
Mon, Nov 07	GHC 6501		available
Fri, Nov 11	GHC 6501		available
Mon, Nov 14	GHC 6501		available
Fri, Nov 18	GHC 6501		SpectraNet: multivariate forecasting and imputation under distribution shifts and missing data Presented by Cristian Challu In this work, we tackle two widespread challenges in real applications for time-series forecasting that have been largely understudied: distribution shifts and missing data. We propose SpectraNet, a novel multivariate time-series forecasting model that dynamically infers a latent space spectral decomposition to capture current temporal dynamics and correlations on the recent observed history. A Convolution Neural Network maps the learned representation by sequentially mixing its components and refining the output. Our proposed approach can simultaneously produce forecasts and interpolate past observations and can, therefore, greatly simplify production systems by unifying imputation and forecasting tasks into a single model. SpectraNet achieves SoTA performance simultaneously on both tasks on five benchmark datasets, compared to forecasting and imputation models, with up to 92% fewer parameters and comparable training times. On settings with up to 80% missing data, SpectraNet has average performance improvements of almost 50% over the second-best alternative.
Mon, Nov 21	GHC 6501		Snapshotting Concurrent Data Structures Presented by Hao Wei Abstract: Most work on concurrent data structures has focused on supporting single-point operations, such as insert, delete, and lookup, but many applications require supporting these alongside multi-point queries, such as searching for ranges of keys, finding the first key that matches some criteria, or checking if a collection of keys are all present. In this presentation, I'll cover a general technique for adding linearizable multi-point queries to existing concurrent data structures. This technique maintains the time bounds and progress properties (e.g. lock-freedom/wait-freedom) of the original data structure. Furthermore, multi-point queries written with this technique are wait-free and take time proportional to their sequential complexity plus a contention term representing the number of update operations concurrent with the query.
Mon, Nov 21	GHC 6501		available
Fri, Nov 25	GHC 6501		available
Mon, Nov 28	GHC 6501		available
Fri, Dec 02	GHC 6501		available
Mon, Dec 05	GHC 6501		available
Fri, Dec 09	GHC 6501		available
Mon, Dec 12	GHC 6501		available
Fri, Dec 16	GHC 6501		available
Mon, Dec 19	GHC 6501		available
Fri, Dec 23	GHC 6501		available

General Info

The Student Seminar Series is an informal seminar for SCS graduate students to communicate ideas. Each meeting starts at noon and lasts 1 hour. Lunch is provided by the Computer Science Department (thanks to Debbie Cavlovich!). At each meeting, a different student speaker will give an informal, 40-minute talk about his/her research, followed by questions/suggestions/brainstorming. We try to attract people with a diverse set of interests, and encourage speakers to present at a very general, accessible level.

So why are we doing this and why take part? In the best case scenario, this will lead to some interesting cross-disciplinary work among people in different fields and people may get some new ideas about their research. In the worst case scenario, a few people will practice their public speaking and the rest get together for a free lunch.

SSS Coordinators

Juncheng Yang, CSD