BEGIN:VCALENDAR VERSION:2.0 PRODID:researchseminars.org CALSCALE:GREGORIAN X-WR-CALNAME:researchseminars.org BEGIN:VEVENT SUMMARY:YUEXIA LUNA LIN (Harvard University) DTSTART:20200710T160000Z DTEND:20200710T170000Z DTSTAMP:20260422T212902Z UID:CRIBB/1 DESCRIPTION:Title: Re ference map technique: a fully Eulerian method for fluid-structure interac tions\nby YUEXIA LUNA LIN (Harvard University) as part of Computationa l Research in Boston and Beyond Seminar (CRIBB)\n\n\nAbstract\nABSTRACT:\n \nFluid-structure interactions (FSI) are abundantly observed in contexts r anging from swimming in the pool\, to industrial level manufacturing\, to bacteria collective motion on a petri dish. However\, the governing equat ions are only analytically trackable in the simple cases\, making simulati ons key to understand this fantastic class of problems. Conventional comp utational methods often create a dilemma for fluid-structure interaction ( FSI) problems. Typically\, solids are simulated using a Lagrangian approa ch with a grid that moves with the material\, whereas fluids are simulated using an Eulerian approach with a fixed spatial grid. FSI methods often r equire some type of interfacial coupling between the two different perspec tives. We present a fully Eulerian FSI method that addresses these challe nges. The method makes use of reference map\, which maps the solid in the current space to the reference space. Reference map is a common concept i n finite strain theory\, but it has been under-utilized as a primary varia ble for solid and FSI simulations. A challenge in applying the reference map technique (RMT) in FSI is to extrapolate reference map values from gri d cells occupied by the solids to unoccupied grid cells\, in order to calc ulate derivative using finite difference schemes. This challenge becomes more acute when applying RMT to simulations in 3D. We develop an extrapol ation algorithm based on least-square linear regression that is suitable f or parallelization. We show examples to demonstrate that RMT is well suit ed for simulating soft\, highly-deformable materials and many-body contact problems. Joint work with Nicholas Derr and Chris H. Rycroft (SEAS\, Har vard University) and Ken Kamrin (Mechanical Engineering\, MIT).\n\nZOOM:\n \nhttps://mit.zoom.us/j/96034732289\n Meeting ID: 960 3473 2289\n Password: 567284\n\n One tap mobile\n +16465588656 \,\,96034732289# US (New York)\n +16699006833\,\,96034732289# US ( San Jose)\n US : +1 646 558 8656 or +1 669 900 6833\n LOCATION:https://researchseminars.org/talk/CRIBB/1/ END:VEVENT BEGIN:VEVENT SUMMARY:Jenelle Feather (MIT) DTSTART:20201002T160000Z DTEND:20201002T050000Z DTSTAMP:20260422T212902Z UID:CRIBB/2 DESCRIPTION:Title: Me tamers of neural networks reveal divergence from human perceptual systems< /a>\nby Jenelle Feather (MIT) as part of Computational Research in Boston and Beyond Seminar (CRIBB)\n\n\nAbstract\nArtificial neural networks now a chieve human-level performance on tasks such as image and speech recogniti on\, raising the question of whether they should be taken seriously as mod els of biological sensory systems. Such neural network models exhibit huma n-like patterns of behavior\, and their feature spaces reliably predict br ain activity. On the other hand\, neural network models can often be foole d by small adversarial perturbations that have no effect on humans. In thi s talk\, I will detail our work using “model metamers” to investigate similarities between neural networks and human sensory systems. Model meta mers are physically distinct stimuli that produce nearly the same response within a model\, and thus the same model prediction. Our results show tha t despite replicating aspects of human behavior and neural responses\, pre sent-day deep neural networks learn invariances that deviate markedly from those of biological sensory systems. Model metamers may help guide future model refinements to reduce or eliminate these discrepancies.\n\nhttps:// mit.zoom.us/j/96155042770 -- Meeting ID: 961 5504 2770\n LOCATION:https://researchseminars.org/talk/CRIBB/2/ END:VEVENT BEGIN:VEVENT SUMMARY:Emily Crabb (MIT) DTSTART:20201106T170000Z DTEND:20201106T180000Z DTSTAMP:20260422T212902Z UID:CRIBB/3 DESCRIPTION:Title: Im portance Of Equilibration Method and Sampling for Ab Initio Molecular Dyna mics Simulations of Solvent - Lithium Salt Systems in Lithium-Oxygen Batte ries\nby Emily Crabb (MIT) as part of Computational Research in Boston and Beyond Seminar (CRIBB)\n\n\nAbstract\nZOOM Link: \n \nhttps: //mit.zoom.us/j/96155042770 - Meeting ID: 961 5504 2770\n\n============= ==============================================================\n\nLithium- oxygen batteries are an active area of research because of their\npotentia l to have a much higher energy density than traditional lithium-ion\nbatte ries. However\, they are not yet commercially viable due to poor\nefficie ncy\, high charging voltages\, and low cycle lifetimes. Many of these\nis sues could be addressed with a deeper fundamental understanding of the\nat omistic behavior of these batteries. One tool to model such atomic scale\ nbehavior is ab initio molecular dynamics (AIMD) simulations. However\, AI MD\nsimulations are limited to timescales of tens of picoseconds due to th eir high\ncomputational cost. As a result\, equilibration and sampling me thodologies can\nhave a significant effect on the behavior of AIMD simulat ions. We thus\ncompared two equilibration methods for AIMD simulations of systems of common\nsolvents and salts found in lithium air batteries: (1) using an AIMD\ntemperature ramp and (2) using a classical MD simulation f ollowed by a short\nAIMD simulation all at the target simulation temperatu re of 300 K. We also\ncompared two different classical all-atom force fie lds (PCFF+ and OPLS) and\nperformed multiple simulations for each system. In this talk\, I will discuss\nwhy lithium-oxygen batteries are an exciti ng area of research\, why\ncomputational tools such as AIMD are critical t o this field\, and how the\ndifferences between our simulation results and experimental results for\nproperties such as coordination number illustra te the importance of both\nequilibration method and independent sampling f or extracting experimentally\nrelevant quantities from AIMD simulations\, with applications in battery \ndevelopment and beyond.\n LOCATION:https://researchseminars.org/talk/CRIBB/3/ END:VEVENT BEGIN:VEVENT SUMMARY:MIRIAM KREHER (MIT) DTSTART:20210205T170000Z DTEND:20210205T180000Z DTSTAMP:20260422T212902Z UID:CRIBB/4 DESCRIPTION:Title: Co mputational Analysis of Nuclear Reactor Transients\nby MIRIAM KREHER ( MIT) as part of Computational Research in Boston and Beyond Seminar (CRIBB )\n\n\nAbstract\nZOOM MEETING info:\n\n https:// mit.zoom.us/j/96155042770\n\n Meeting ID: 9 61 5504 2770\n\n\nSince nuclear experiments are costly and require extensi ve safety precautions\, the nuclear industry relies heavily on modeling an d simulation of nuclear systems. The state-of-the-art simulation tool for steady-state neutron transport is Monte Carlo\, a probabilistic approach to solving for the distribution of neutrons. \n\nAlthough it is the most a ccurate tool available\, it is very computationally expensive. Monte Carlo is even more burdensome when coupled to other physics which allows us to properly capture feedback effects from density and temperature changes. No netheless\, it is imperative to do such coupling because nuclear reactor d esigns rely on these intrinsic feedback mechanisms to ensure passive safet y. In addition to coupling Monte Carlo with other physics codes\, there is an additional hurdle to overcome for time-dependent simulations. These a re a few of the reasons why nuclear reactor simulations are a target of Ex ascale computing initiatives. \n\nThis talk will cover a number of couplin g schemes that create feasible runtimes for coupled time-dependent Monte C arlo simulations. In particular\, we will give consideration to high-orde r/low-order schemes where Monte Carlo and diffusion solvers are paired to deliver accurate results in efficient time. \n\n\nABOUT THE SPEAKER: Miri am Kreher is a PhD candidate in the Computational Reactor Physics Group in the MIT Nuclear Science and Engineering Department. She is also a fellow of the DOE Computational Science Graduate Fellowship program. Kreher rece ived a BS in Engineering Science from the University of Pittsburgh in 2016 . Kreher is a contributor of OpenMC and currently serves on the Board of D irectors of the American Nuclear Society.\n LOCATION:https://researchseminars.org/talk/CRIBB/4/ END:VEVENT BEGIN:VEVENT SUMMARY:Kevin Silver (MIT) DTSTART:20210305T170000Z DTEND:20210305T180000Z DTSTAMP:20260422T212902Z UID:CRIBB/5 DESCRIPTION:Title: Bu ckling\, Crumpling\, And Tumbling Of Semiflexible Sheets In Simple Shear F low\nby Kevin Silver (MIT) as part of Computational Research in Boston and Beyond Seminar (CRIBB)\n\n\nAbstract\nAs 2D materials such as graphen e\, transition metal dichalcogenides\, and 2D polymers become more prevale nt\, solution processing and colloidal-state properties are being exploite d to create advanced and functional materials. However\, our understandin g of the fundamental behavior of 2D sheets and membranes in fluid flow is still lacking. In this work\, we perform numerical simulations of atherma l semiflexible sheets with hydrodynamic interactions in shear flow. For s heets initially oriented in the flow-gradient plane\, we find buckling ins tabilities of different mode numbers that vary with bending stiffness and can be understood with a quasi-static model of elasticity. For different i nitial orientations\, chaotic tumbling trajectories are observed.\n\n ZOOM MEETING info:\n\n https://mit.zoom.us/j/96155042770\n\n M eeting ID: 961 5504 2770\n LOCATION:https://researchseminars.org/talk/CRIBB/5/ END:VEVENT BEGIN:VEVENT SUMMARY:Siddharth Samsi and Vijay Gadepally (MIT-Lincoln Lab) DTSTART:20210402T160000Z DTEND:20210402T170000Z DTSTAMP:20260422T212902Z UID:CRIBB/6 DESCRIPTION:Title: An Open Datacenter Dataset for AI Enabled Optimization\nby Siddharth Sam si and Vijay Gadepally (MIT-Lincoln Lab) as part of Computational Research in Boston and Beyond Seminar (CRIBB)\n\n\nAbstract\nThe first step in tra ining an AI is to get the right data. In order to apply AI to the problem of data center optimization\, such as identifying faults with servers\, e nergy or cooling systems\, before they become critical\, the MIT Lincoln L aboratory Supercomputing Center is developing a state-of-the-art dataset. This dataset contains rich information such as: physical information abou t building management\; system information such as scheduler and filesyste m logs\; and node-level information such as utilization\, memory\, GPU act ivity (both job level statistics as well as time-series monitoring collect ed via NVIDIA’s DCGM tool)\, energy utilization\, etc. In this talk\, we will describe the dataset\, detail how developers can get access to this data\, and discuss a number of open problems associated with datacenter an alytics.\n LOCATION:https://researchseminars.org/talk/CRIBB/6/ END:VEVENT BEGIN:VEVENT SUMMARY:Peter James Ahrens (MIT) DTSTART:20210507T160000Z DTEND:20210507T170000Z DTSTAMP:20260422T212902Z UID:CRIBB/7 DESCRIPTION:Title: On Optimal Partitioning for Variable Block Row Format\nby Peter James Ah rens (MIT) as part of Computational Research in Boston and Beyond Seminar (CRIBB)\n\nLecture held in Virtual.\n\nAbstract\nThe Variable Block Row (V BR) format is an influential blocked sparse matrix format designed for mat rices with shared sparsity structure between adjacent rows and columns. VB R groups adjacent rows and columns\, storing the resulting blocks that con tain nonzeros in a dense format. This reduces the memory footprint and en ables optimizations such as register blocking and instruction-level parall elism. Existing approaches use heuristics to determine which rows and col umns should be grouped together. We show that finding the optimal groupin g of rows and columns for VBR is NP-hard under several reasonable cost mod els. In light of this finding\, we propose a 1-dimensional variant of VBR\ , called 1D-VBR\, which achieves better performance than VBR by only group ing rows. We describe detailed cost models for runtime and memory consump tion. Then\, we describe a linear time dynamic programming solution for o ptimally grouping the rows for 1D-VBR format. We extend our algorithm to p roduce a heuristic VBR partitioner which alternates between optimally part itioning rows and columns\, assuming the columns or rows to be fixed\, res pectively. Our alternating heuristic produces VBR matrices with the smalle st memory footprint of any partitioner we tested.\n\n ZOOM MEETING info:\n\n https:/ /mit.zoom.us/j/96155042770\n\n Meeting ID : 961 5504 2770\n LOCATION:https://researchseminars.org/talk/CRIBB/7/ END:VEVENT BEGIN:VEVENT SUMMARY:Kurt Keville (MIT) DTSTART:20210604T160000Z DTEND:20210604T170000Z DTSTAMP:20260422T212902Z UID:CRIBB/8 DESCRIPTION:Title: Bi g Memory servers and modern approaches to disk-based computation\nby K urt Keville (MIT) as part of Computational Research in Boston and Beyond S eminar (CRIBB)\n\n\nAbstract\nThere is a new computing paradigm available today facilitated by commodity server platforms. It is often called Big Me mory solutions because it exposes a large RAM subsystem to the Operating S ystem and therefore afford the application programmer a number of previous ly unavailable options for data management. Additionally\, certain vendor- specific solutions offer additional memory management options that pay div idends in data reliability and access speeds. A survey of these offerings and the promise of massive memory compute will be discussed.\n\n ZOOM MEETING info:\n\n https://mit.zoom.us/j/96155042770\n\n Meeting ID: 961 5504 2770\n LOCATION:https://researchseminars.org/talk/CRIBB/8/ END:VEVENT BEGIN:VEVENT SUMMARY:Steven Torrisi (Harvard University) DTSTART:20210723T160000Z DTEND:20210723T170000Z DTSTAMP:20260422T212902Z UID:CRIBB/9 DESCRIPTION:Title: Wh ich parts matter? Interpretable random forest models for X-Ray absorption spectra\nby Steven Torrisi (Harvard University) as part of Computatio nal Research in Boston and Beyond Seminar (CRIBB)\n\n\nAbstract\nX-ray abs orption spectroscopy (XAS) produces a wealth of information about the loca l structure of materials\, but interpretation of spectra often relies on e asily accessible trends and prior assumptions about the structure. Recentl y\, researchers have demonstrated that machine learning models can automat e this process to model the environments of absorbing atoms from their XAS spectra. However\, machine learning models are often difficult to interpr et\, making it challenging to determine when they are valid and whether th ey are consistent with physical theories. In this work\, we present three main advances to the data-driven analysis of XAS spectra: we demonstrate t he efficacy of random forests in solving two new property determination ta sks (predicting Bader charge and mean nearest neighbor distance)\, we addr ess how choices in data representation affect model interpretability and a ccuracy\, and we show that multiscale featurization can elucidate the regi ons and trends in spectra that encode various local properties. The multis cale featurization transforms the spectrum into a vector of polynomial-fit features\, and is contrasted with the commonly-used “pointwise” featu rization that directly uses the entire spectrum as input. We find that acr oss thousands of transition metal oxide spectra\, the relative importance of features describing the curvature of the spectrum can be localized to i ndividual energy ranges\, and we can separate the importance of constant\, linear\, quadratic\, and cubic trends\, as well as the white line energy. \n\nThis work has the potential to assist rigorous theoretical interpreta tions\, expedite experimental data collection\, and automate analysis of X AS spectra\, thus accelerating the discovery of new functional materials. We expect that this featurization strategy could be useful for broad domai ns of application\, such as one-dimensional time-series analysis or other forms of spectroscopy.\n\nPaper: https://www.nature.com/articles/s41524-02 0-00376-6\n\n============================================================= ==============\n\nZOOM MEETING info:\n\n https://mit.zoom.us/j /96155042770\n\n Meeting ID: 961 5504 2770\n LOCATION:https://researchseminars.org/talk/CRIBB/9/ END:VEVENT BEGIN:VEVENT SUMMARY:Jenny Coulter (Harvard University) DTSTART:20210806T160000Z DTEND:20210806T170000Z DTSTAMP:20260422T212902Z UID:CRIBB/10 DESCRIPTION:Title: P hoebe: A New Open-Source Package for Electrical and Thermal Materials Tran sport Predictions from First-Principles\nby Jenny Coulter (Harvard Uni versity) as part of Computational Research in Boston and Beyond Seminar (C RIBB)\n\n\nAbstract\nUnderstanding the electrical and thermal transport pr operties of materials is critical to the design of all kinds of devices. T he theoretical prediction of these quantities relies on an accurate descri ption of the electron and phonon properties of each material. Additionally \, a number of different quasiparticle interactions must be considered to accurately predict transport behavior. While first-principles methods base d on density functional theory can describe these material-specific quasip article properties\, using this information to calculate transport coeffic ients can be computationally demanding and memory intensive. \n\nTo addres s this challenge\, we present a recently developed software package\, Phoe be (https://github.com/mir-group/phoebe)\, which includes the effects of e lectron-phonon\, phonon-phonon\, boundary\, and isotope scattering to pred ict the electron and phonon transport properties of materials by solving t he Boltzmann transport equation (BTE) using a scattering matrix formalism. This open source C++ code utilizes MPI-OpenMP hybrid parallelization as w ell as GPU acceleration and distributed memory structures to manage comput ational cost and take advantage of modern HPC systems. Using this new fram ework\, we are able to accurately and efficiently predict a wide range of material transport properties such as the electrical and thermal conductiv ity and thermoelectric performance. \n\n ht tps://math.mit.edu/sites/crib/\n\n Z OOM MEETING info:\n\n https://mit.zoom.us/j/ 96155042770\n Meeting ID: 961 5504 2770 \n LOCATION:https://researchseminars.org/talk/CRIBB/10/ END:VEVENT END:VCALENDAR