150813 research outputs found
Sort by
Report to the President for year ended June 30, 2025, Program in Media Arts and Sciences
No full textThis report contains the following sections: Graduate Program, Undergraduate Engagement, Visiting Students, Faculty and Staff, Diversity & Student Support, MIT Summer Research Program (MSRP), Open House, SOS (Students Offering Support), Recruitment Update, Masters Enrollment (2015-2025), Secondary Advisor Program, MAS Seminar Series, Cultivating Creativity, Constructionist Learning, and/or Human Flourishing, Student Research Seminar Series, MAS Alumni Mentorship Program, Career and Professional Development Series, Alumni Career Connect, Creative Group Meetup, International Student Meet & Greet Lunch with Sylvia Hiestand, Identity Student Support Lunches, Graduate Reflection Lunches, Community-Building Events:, Study Breaks, MAS Student Summer Lunch Series, Crafting Events, Halloween Trick-or-Treating, Community Appreciation Events: Celebrating Kindness and Graduate Contributions, Ice Cream Social, Women Take The Reel Film Festival Screenings, PRIDE Community Lunch, MAS Commencement Celebration, Diversity, Equity, Inclusion, Learning & Student Support Workshop Series, and Sampling of Honors and Awards
On efficiently computable functions, deep networks and sparse compositionality
No full textIn previous papers [4, 6] we have claimed that for each function which is efficiently Turing computable
there exists a deep and sparse network which approximates it arbitrarily well. We also claimed a key role
for compositional sparsity in this result. Though the general claims are correct some of our statements
may have been imprecise and thus potentially misleading. In this short paper we wish to formally restate
our claims and provide definitions and proofs.This work was supported by the Center for Brains, Minds and Machines (CBMM), funded by NSF STC award CCF-1231216
Moderating Large Scale Online Deliberative Processes with Large Language Models (LLMs): Enhancing Collective Decision-Making.
No full textSAC ’25, March 31-April 4, 2025, Catania, ItalyThis study investigates the use of LLMs, specifically ChatGPT-4o, to enhance the moderation of online deliberative processes. Traditionally, decision-making has been controlled by small groups, often excluding the vital insights that crowd intelligence can provide. As global challenges grow more complex, broader and more inclusive participation is essential. While online platforms allow for such large-scale participation, they also face significant issues, including content fragmentation, low signal-to-noise ratios, and inefficient argumentation. Human moderators can address these challenges, but scaling them is prohibitively costly. This research introduces a more scalable solution by leveraging LLMs to automate critical moderation tasks, including unbundling multiple ideas, categorizing them into solutions, metrics, and barriers, and implementing efficient argument mining and classification techniques. Additionally, it evaluates the effectiveness of different prompting styles in optimizing moderation. The findings demonstrate that LLMs can successfully moderate key aspects of large-scale online deliberations, such as unbundling and categorization, improving the structure of discussions and representing a significant step forward in collective decision-making
Higher rank flag sheaves on surfaces
No full textWe study moduli space of holomorphic triples E 1 → ϕ E 2 , composed of torsion-free sheaves E i , i = 1 , 2 , and a holomorphic mophism between them, over a smooth complex projective surface S. The triples are equipped with Schmitt stability condition (Schmitt in Algebras Represent Theory 6(1):1–32, 2000). We observe that when Schmitt stability parameter q(m) becomes sufficiently large, the moduli space of triples benefits from having a perfect relative and absolute deformation-obstruction theory in some cases. We further generalize our construction by gluing triple moduli spaces, and extend the earlier work (Gholampour et al. in Nested Hilbert schemes on surfaces: virtual fundamental class, preprint, arXiv:1701.08899 ) where the obstruction theory of nested Hilbert schemes over the surface was studied. Here we extend the earlier results to the moduli space of chains E 1 → ϕ 1 E 2 → ϕ 2 ⋯ → ϕ n - 1 E n , where ϕ i are injective morphisms and rk ( E i ) ⩾ 1 for all i. There is a connection, by wallcrossing in the master space, between the theory of such higher rank flags, and the theory of Higgs pairs on the surface, which provides the means to relate the flag invariants to the local DT invariants of threefold given by a line bundle on the surface, X := Tot(L → S)
Effects of surface species and homogeneous reactions on rates and selectivity in ethane oxidation on oxide catalysts
No full textSelective alkane oxidations on metal oxide catalysts involve complex mechanisms with multiple reactions in series and parallel, different types of reduced and oxidized surface species, and potential contributions from gas-phase reactions. Here, kinetics and thermodynamics of elementary steps involved in C2H6-O2 reactions on SiO2-supported small vanadium oxide domains are determined using density functional theory. These surface reactions together with gas-phase mechanisms are incorporated in kinetic simulations to determine how surface and gaseous reactions interact and contribute to rates and selectivity. The results show that gas-phase reactions within pore volumes in contact with the catalyst contribute significantly to C2H6 activation rates, even at conditions where gas-phase reactions in empty volumes without catalyst are negligible. The majority of C2H6 activations occur on the surface, via H abstraction by vanadium oxo species present at terminal lattice oxygens. The gas-phase activations via H-abstraction by OH radicals also exhibit significant contributions. The reduced centers formed by reactions at vanadium oxo species are re-oxidized rapidly and, therefore, are present in very small concentrations at reaction conditions. The re-oxidation steps lead to the formation of HO2 radicals and surface peroxo species that are also rapidly consumed and are present in small concentrations. The peroxo species preferentially convert C2H4 to its epoxide product and influence selectivity even at low concentrations. The gas-phase reactions decrease the concentrations of peroxo species and improve selectivity slightly. The effects of reaction conditions and catalyst site density provide further insights into how factors beyond conversions at lattice oxygens influence rates and selectivity in alkane oxidation reactions of significant industrial importance
FluidFlower: A Meter-Scale Experimental Laboratory for Geological CO2 Storage
No full textThe original idea of constructing the FluidFlower was to construct an experimental laboratory that was well suited to both scientific research and public outreach. Indeed, a core principle was to allow for demonstrating the key physical mechanisms underpinning geological CO2 storage to the public in what can be perceived as a realistic setting. This motivated the design of a relatively large experiment (about 3 by 2 m), with a transparent glass plate, and where pH sensitive dye was used to mark the CO2 concentration in the water phase. With these dimensions, some geological complexity could be included in the experiment, and the use of high-permeable unconsolidated sands reduced the timescales to hours and days, as opposed to the years and centuries of relevance at field conditions.
The science part of the FluidFlower study was facilitated by the serendipitous arrival of the Covid-19 pandemic. We realized that the construction of the FluidFlower was at a scale and purpose which was quite unique, and that the travel restrictions imposed by Covid-19 allowed us to limit the insight non-local scientists would have in the experiments we conducted. This motivated the design of, and call for participation in, a forecasting study during spring 2021—and to our great fortune, good colleagues from around the globe agreed to participate.
The main part of the study took place from early fall 2021 through April 2022, and during this process, it quickly became clear that there was much more to be said about this study than what could fit within a single paper. The idea for creating the special issue you are now reading was thus formed
Fundamentals, Voltage Control and Novel Application of Exchange Bias in Magnetic Thin Films
No full textThe past half-century has seen remarkable advances in microelectronics, but as transistors approach their physical limits, there is a growing need for beyond-CMOS technologies. Spintronics, which aims to utilize the electron's spin in magnetic thin films for data storage and manipulation, is a promising alternative. Among the rich physical interactions that appear in magnetic thin films, the exchange-bias (EB) effect is essential for many spintronic devices. EB is an effect that arises at ferromagnet/antiferromagnet interfaces, which imposes an internal field on the ferromagnet, enhancing the range of functionalities that can be derived from devices. This thesis explores EB in Co/Co₁₋ₓNiₓO systems at multiple levels, from fundamental understanding to its manipulation and applications. First, we introduce a new model to predict EB in polycrystalline antiferromagnetic thin films and validate it with experimental data. Second, we tackle another fundamental aspect – disentangling the effects of EB on nucleation and propagation of magnetization reversal. We discover that nucleation and propagation EB can be unequal and demonstrate how that can lead to unexpected behavior of the system, including having asymmetric hysteresis loops. Building on these insights, we demonstrate voltage-controlled ionic gating to manipulate EB, achieving cyclic toggling of the EB sign in a ferrimagnetic system, where the magnetization direction is fully determined by the gating state. Furthermore, by targeting the antiferromagnet directly, we discover EB enhancement up to 100%, which can be explained with the help of the model developed earlier. We demonstrate sub-millisecond and analog operation in this system. Finally, a new approach to improving bit-stability whilst preserving performance in magnetic racetrack memory is proposed which involves incorporating an EB layer with the right properties into the track. The benefits obtained from this strategy can help push this next-gen memory device closer to commercialization. We believe the findings in this thesis substantially extend the state-of-the-art in terms of basic understanding of EB, ways of EB manipulation and unexplored use-cases of EB, paving the way for new functionalities in spintronic devices applicable for non-conventional computing paradigms or next-gen memory devices.Ph.D
Measurement of the inclusive t t ¯ cross section in final states with at least one lepton and additional jets with 302 pb−1 of pp collisions at s = 5.02 TeV
No full textA measurement of the top quark pair ( t t ¯ ) production cross section in proton-proton collisions at a centre-of-mass energy of 5.02 TeV is presented. The data were collected at the LHC in autumn 2017, in dedicated runs with low-energy and low-intensity conditions with respect to the default configuration, and correspond to an integrated luminosity of 302 pb−1. The measurement is performed using events with one electron or muon, and multiple jets, at least one of them being identified as originating from a b quark (b tagged). Events are classified based on the number of all reconstructed jets and of b-tagged jets. Multivariate analysis techniques are used to enhance the separation between the signal and backgrounds. The measured cross section is 62.5 ± 1.6 stat − 2.5 + 2.6 syst ± 1.2 lumi pb. A combination with the result in the dilepton channel based on the same data set yields a value of 62.3 ± 1.5 (stat) ± 2.4 (syst) ± 1.2 (lumi) pb, to be compared with the standard model prediction of 69.5 − 3.7 + 3.5 pb at next-to-next-to-leading order in perturbative quantum chromodynamics
Affine Springer Fibers and the Kazhdan-Lusztig Map
No full textLet G be a connected reductive group with Lie algebra g and Weyl group W. Let P ⊂ G((t)) be a parahoric subgroup with Levi quotient Gₚ. Using the topology of Lie P, Kazhdan and Lusztig define a map from nilpotent orbits in Lie Gₚ to conjugacy classes in W. This thesis proves compatibilities between Kazhdan-Lusztig maps associated to different parahoric subgroups, as well as the Kazhdan-Lusztig map for the Langlands dual. These compatibilities come from studying the W-representation on the cohomology of affine Springer fibers. The main tool is Yun’s Global Springer Theory. We give two applications of these compatibilities. The first is an affine analog of the classical picture relating singular supports of IC sheaves on the flag variety with special nilpotent orbits. The second is a resolution of Lusztig’s conjecture that strata can be described by fibers of (parahoric) Kazhdan-Lusztig maps.Ph.D
Chemprop: A Machine Learning Package for Chemical Property Prediction
No full textDeep learning has become a powerful and frequently employed tool for the prediction of molecular properties, thus creating a need for open-source and versatile software solutions that can be operated by nonexperts. Among the current approaches, directed message-passing neural networks (D-MPNNs) have proven to perform well on a variety of property prediction tasks. The software package Chemprop implements the D-MPNN architecture and offers simple, easy, and fast access to machine-learned molecular properties. Compared to its initial version, we present a multitude of new Chemprop functionalities such as the support of multimolecule properties, reactions, atom/bond-level properties, and spectra. Further, we incorporate various uncertainty quantification and calibration methods along with related metrics as well as pretraining and transfer learning workflows, improved hyperparameter optimization, and other customization options concerning loss functions or atom/bond features. We benchmark D-MPNN models trained using Chemprop with the new reaction, atom-level, and spectra functionality on a variety of property prediction data sets, including MoleculeNet and SAMPL, and observe state-of-the-art performance on the prediction of water-octanol partition coefficients, reaction barrier heights, atomic partial charges, and absorption spectra. Chemprop enables out-of-the-box training of D-MPNN models for a variety of problem settings in fast, user-friendly, and open-source software