BEGIN:VCALENDAR VERSION:2.0 X-WR-CALNAME:EventsCalendar PRODID:-//hacksw/handcal//NONSGML v1.0//EN CALSCALE:GREGORIAN BEGIN:VTIMEZONE TZID:America/New_York LAST-MODIFIED:20240422T053451Z TZURL:https://www.tzurl.org/zoneinfo-outlook/America/New_York X-LIC-LOCATION:America/New_York BEGIN:DAYLIGHT TZNAME:EDT TZOFFSETFROM:-0500 TZOFFSETTO:-0400 DTSTART:19700308T020000 RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU END:DAYLIGHT BEGIN:STANDARD TZNAME:EST TZOFFSETFROM:-0400 TZOFFSETTO:-0500 DTSTART:19701101T020000 RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU END:STANDARD END:VTIMEZONE BEGIN:VEVENT CATEGORIES:College of Arts and Sciences,Lectures and Seminars DESCRIPTION:Abstract: Time series forecasting is essential across domains s uch as healthcare, energy, and climate modeling. While models like LSTMs, GRUs, Transformers, and State-Space Models (SSMs) have become widely used, selecting the optimal architecture remains unclear. We propose an automat ed framework that systematically designs hybrid architectures by combining LSTM, GRU, attention, and SSM modules. Our approach uses multi-objective optimization to explore combinations and orderings of blocks, yielding Par eto-optimal architectures that balance user-defined trade-offs among objec tives. A preference function selects the most suitable model for a given a pplication. Moreover, two sampling-based iterative procedures for Pareto-f ront exploration are introduced, which reduces the total training cost by nearly eightfold. Across four real-world benchmarks, our framework reveals that simple models excel in speed, while hybrid compositions dominate whe n balancing accuracy and complexity. Our findings challenge the notion of a universally superior neural architecture, emphasizing instead the value of data- and objective-driven design in time series forecasting.\nEvent pa ge: /events/cms/cscdr-seminar-by-qianying-cao-brown- on-automatic-selection-of-the-best-neural-architecture-for-time-series-for ecasting.php\nEvent link: https://www.cscdr.umassd.edu/seminars X-ALT-DESC;FMTTYPE=text/html:

糖心logo入口

Abstract:

Time series f orecasting is essential across domains such as healthcare\, energy\, and c limate modeling. While models like LSTMs\, GRUs\, Transformers\, and State -Space Models (SSMs) have become widely used\, selecting the optimal archi tecture remains unclear. We propose an automated framework that systematic ally designs hybrid architectures by combining LSTM\, GRU\, attention\, an d SSM modules. Our approach uses multi-objective optimization to explore c ombinations and orderings of blocks\, yielding Pareto-optimal architecture s that balance user-defined trade-offs among objectives. A preference func tion selects the most suitable model for a given application. Moreover\, t wo sampling-based iterative procedures for Pareto-front exploration are in troduced\, which reduces the total training cost by nearly eightfold. Acro ss four real-world benchmarks\, our framework reveals that simple models e xcel in speed\, while hybrid compositions dominate when balancing accuracy and complexity. Our findings challenge the notion of a universally superi or neural architecture\, emphasizing instead the value of data- and object ive-driven design in time series forecasting.

Event page: /events/cms/cscdr-seminar-by-qianying-cao-b rown-on-automatic-selection-of-the-best-neural-architecture-for-time-serie s-forecasting.php
Event link:

DTSTAMP:20260410T072746 DTSTART;TZID=America/New_York:20260408T130000 DTEND;TZID=America/New_York:20260408T140000 LOCATION:Textiles 105A SUMMARY;LANGUAGE=en-us:CSCDR Seminar by Qianying Cao (Brown) on Automatic s election of the best neural architecture for time series forecasting UID:d98f81eef1baef5233b2c3c3b7fd106f@www.umassd.edu END:VEVENT END:VCALENDAR