146 research outputs found
TACO library interface and runtime
No full textThis electronic version was submitted by the student author. The certified thesis is available in the Institute Archives and Special Collections.Thesis: M. Eng., Massachusetts Institute of Technology, Department of Electrical Engineering and Computer Science, 2019Cataloged from student-submitted PDF version of thesis.Includes bibliographical references (pages 65-66 ).Tensor algebra is a powerful tool for computing on multidimensional data and has applications in many fields. The number of possible tensor operations is infinite, so it is impossible to manually implement all of them to operate on different tensor dimensions. The Tensor Algebra Compiler (taco) introduced a compiler approach to automatically generate kernels for any compound tensor algebra operation on any input tensor formats. In this thesis, we present a new API for the taco library. The API removes the need to call compiler methods with the introduction of a delayed execution framework. Additionally, the API introduces multiple important tensor algebra features previously unavailable in taco. Finally, we propose extensions to taco's code generation algorithm to automatically generate tensor API methods for any tensor format. The proposed API makes taco code cleaner, shorter and more elegant. Furthermore, the extensions to its code generation algorithm make the API scalable to new formats and operations.by Patricio Noyola.M. Eng.M.Eng. Massachusetts Institute of Technology, Department of Electrical Engineering and Computer Scienc
Taco Cart Lesson, a Three Act Task (pp. 91--106)
No full textThe author offers a revision of a lesson plan which uses Dan Meyer’s Three Act Task format tolead students through a problem involving the Pythagorean Theorem. The Taco Cart problem uses a reallife situation to engage students. The research-based revisions that the author offers helps increase clarityfor students and teachers.
Taco Tensor Algebra kernels on distributed systems using Legion
No full textThis electronic version was submitted by the student author. The certified thesis is available in the Institute Archives and Special Collections.Thesis: M. Eng., Massachusetts Institute of Technology, Department of Electrical Engineering and Computer Science, 2019Cataloged from student-submitted PDF version of thesis.Includes bibliographical references (pages 89-91).Tensor algebra is a powerful language for expressing computation on multidimensional data. While many tensor datasets are sparse, most tensor algebra libraries have limited support for handling sparsity. The Tensor Algebra Compiler (Taco) has introduced a taxonomy for sparse tensor formats that has allowed them to compile sparse tensor algebra expressions to performant C code, but they have not taken advantage of distributed systems. This work provides a code generation technique for creating Legion programs that distribute the computation of Taco tensor algebra kernels across distributed systems, and a scheduling language for controlling how this distributed computation is structured. This technique is implemented in the form of a command-line tool called SuperTaco. We perform a strong scaling analysis for the SpMV and TTM kernels under a row blocking distribution schedule, and find speedups of 9-10x when using 20 cores on a single node. For multi-node systems using 20 cores per node, SpMV achieves a 33.3x speedup at 160 cores and TTM achieves a 42.0x speedup at 140 cores.by Sachin Dilip Shinde.M. Eng.M.Eng. Massachusetts Institute of Technology, Department of Electrical Engineering and Computer Scienc
Conversations about adulthood recorded in the album "1-800 Enlightenment" by Taco Hemingway - rap as a learning space
No full textRap music is part of hiphop culture, and this music genre is has been developing in Poland since 1990s.The lyrics of rap songs are often biographical in nature, and their creators refer to their own experiences and experiences, which they describe and interpret, giving them subjective meaning. In this text, the author analyzed the work of Taco Hemingway - one of the most popular Polish rappers. She focused on the album titled “1-800 Enlightenment”, which tells the story of an artist overcoming a crisis and at the same time draws attention to the problems of adults. The main motive and message of the artist is to encourage people to talk about difficulties and problems, but also to reflect on their own lives. The author, referring to the biographical content contained in the artist’s album, analyzes the album in terms of its educational potential. It refers to the perspective of informal learning, and in particular to learning from one’s own biography and the biographies of others. The discussion on the artist’s album made it possible to describe rap as a space for adult learning
Elastic registration of histological serial sections: A finite element approach
No full textFor a proper three-dimensional reconstruction of histology serial sections, adjustment of the slices is necessary for combining serial sections. Deformations occur due to the sectioning and acquisition pro-cess of the microscopic analysis of histology. By reconstructing the deformations with a transformation of the sections, a mathematical correction on the images can be applied. By using image registration, a transformation function is searched for to minimize differences between the histology slices. Due to the non-linearity of the distortions, prior knowledge is required in order to have a solvable problem. Additional information in the form of elasticity regularisation is considered. Implementing the elastic regularisation with a finite element method, provides a continuous transformation function With the continuous function, in a natural way, alignment can be monitored for folding transformations. In this work, the (bi-)linear and (bi-)quadratic elements for the finite element method are implemented and compared with the finite difference method. It is observed that for the different kinds of elements, the (bi-)linear elements yield best results with the validity of the transformation. Moreover, the computa- tional costs for the bi-linear elements are the cheapest. Compared with the finite difference method, the differences in accuracy are not noteworthy but the computational time of the finite element method is longer. Furthermore, to steer the matching in an accurate direction, improvements are proposed by applying local stiffness of the elements or adding soft constraints on the volume of the elements. This results in significant improvements in the transformation. For these two approaches it is observed that local stiffness is more restrictive than volume-preserving. Solving the optimization problem, a Gauss-Newton method to search for descent directions is applied. A matrix-based and a matrix-free approach of elasticity regularisation is considered in the linear system of finding a descent direction. While the matrix-free approach decreases the memory usage, the computational costs are significantly increased
Fuel Flexibility: Fuel-flexible fuel cell systems for super yachts
No full textThe reduction of the of harmful emissions that result fromyachting are an increasing concern and priority. It has been widely established that yachts have a significant contribution to the emission of greenhouse gasses and other emissions that are harmful to the environment and the public health. Measures must be taken that drastically reduce the environmental impact of yachts. The use of fuel cells, in combinations with alternative fuels, can provide a solution. The emission of all harmful emissions can be greatly reduced or completely eliminated. One of the major challenges in this transition is the uncertainty in the future supply of alternative fuels. Yachts are built to operate up-to 30 years, this is only possible if it is ensured of fuel supply over this time. Fuel-flexibility can provide a solution. If the operating fuel can be changed according to the availability, the yacht is guaranteed of operation in all future scenarios. This report describes research on how a fuelflexible fuel cell system can be designed that is capable op powering a super yacht. A selection of alternative fuels was made. These fuels must be suitable for marine application, they must hold environmental benefits over conventional fuels, and together they must ensure operation in all future scenarios. A number of fuels was considered. Their storage, production and feedstocks were compared. A final selection of operative fuels was made: it consists of ammonia, methanol, and bio/synthetic diesel. A fuel-flexible fuel cell system was designed based on the power demands of a yacht. The operational demands have been carefully mapped based on a yacht design of De Voogt Naval Architects. The system design is based on an SOFC with an external (pre-)reformer/cracking reactor. This reactor operates adiabatic and the required steam is supplied by water recirculation, both to increase the control over the fuel decomposition process. Waste heat is utilized in the balance of plant of the system, and can be recovered from the system in the form of a hot water flow. This heat can be consumed directly during the operation of the yacht. There are limitations in the fuel-flexibility of the system. Some of the components must be replaced when a switch between alternative fuels is made. A performance analysis is needed to find whether the fuel-flexible fuel cell system can power a yacht. A model of the system was made in which the performance of the system was simulated under the defined operational requirements. A thermodynamic analysis of the system is conducted to find the efficiencies and fuel consumption, and the heat and mass flows within the system. These parameters were used to calculate the yearly fuel consumption, emission of CO2, and required stored fuel capacity. The efficiency for the designed fuel-flexible fuel cell system is high for all of the fuels, and throughout the operation of the yacht. There is a large difference in fuel consumption and required fuel storage capacity between the alternative fuels, this is mainly caused by the energy density differences of the fuels. The emission of CO2 is depended on the fuel consumption and the carbon content of the fuels. Compared to conventional power generation and fossil fuels, the emission of CO2 is significantly reduced. There are also differences in the heat and mass flows within the system. Each fuel therefore has a different required capacity for pieces of equipment such as heat exchangers, pumps, and compressors. This research has shown how a fuel-flexible fuel cell system can be designed that is capable of powering a yacht. Through a thermodynamic model, it is shown that such a system has a high operational efficiency throughout the performance of the yacht. Harmful emissions are greatly reduced compared to conventional power generation. In a fuel-flexible system, each component must be designed for themost demanding fuel. For each fuel, some system components will therefore be over-dimensioned. A fuel-flexible system will be more complex and voluminous than a single-fuel fuel cell system. In return, the system will be operational in all future scenarios. This research has shed light on the trade-offs that the choice for a fuel-flexible fuel cell system can bring about. It provides an analysis which will only become more valuable as the yachting industry is urged to adapt to an ever-changing environmentMarine Technology | Marine Engineerin
Long-term outcomes of children treated for HIV-infection
No full textDespite the drastic decrease of mortality and morbidity since the introduction of combination antiretroviral therapy (cART) in 1996, children with a chronic HIV-infection remain vulnerable to multiple complications that may negatively impact their physical health, cognitive function and psychological wellbeing. The first part of this thesis focuses on HIV-infected children in care in the Netherlands since 1996. The majority of these children are immigrants from sub-Saharan Africa. They had a sustained long-term immunological reconstitution, which was independent of age at cART-initiation. The percentage of treated children with an undetectable HIV viral load rose substantially over time, and mortality rate was low as compared to other industrialized countries. Part 2 addresses various complications of chronic HIV-infection in children. We found that the cognitive performance of HIV-infected children was poorer as compared to a group of age-, gender-, ethnicity- and socioeconomically-matched controls. Neuroimaging showed that the HIV-infected group had a lower cortical and white matter volume, more white matter hyperintensities and poorer white matter integrity. A thinner retinal fovea was found in the HIV-infected children, however visual function was not affected. No decline in health-related quality of life was found. Lastly, we showed that subcutaneous fat loss is still prevalent in children treated for HIV, and is strongly associated with the antiretroviral drug stavudine; which although eliminated from treatment protocols is still often prescribed in developing countries where alternatives are sparse. Future research in HIV-infected children should focus on pathophysiology behind the detected complications, and ultimately lead to their treatment and prevention
Google in China
No full textThe Mekelprize 2007 was won by Taco Broerse, who wrote an essay titled Google in China. Is it morally justified for a Western company to go along with self-censorship, especially in a country like China, where freedom of expression is under pressure? The jury found this essay to provide a strong analysis. The central question is explored from different angles and well supported. In addition, this student shows himself to be a real engineer-to-be by not only giving an analysis, but also trying to come up with directions for solutions.Civil Engineering and Geoscience
Group theory and information theory algorithms in deep learning
No full textSubmission original under an indefinite embargo labeled 'Open Access'. The submission was exported from vireo on 2025-02-04 without embargo termsThe student, Sourya Basu, accepted the attached license on 2024-07-10 at 15:02.The student, Sourya Basu, submitted this Dissertation for approval on 2024-07-10 at 15:15.This Dissertation was approved for publication on 2024-07-11 at 08:47.DSpace SAF Submission Ingestion Package generated from Vireo submission #21048 on 2025-02-04 at 21:04:54We are witnessing an enormous growth in deep learning that impacts our day-to-day lives with applications ranging from drug discovery to coding assistants. Accelerating this growth requires optimizing the several stages of building deep learning models such as designing efficient architectures, fine-tuning pretrained models, and sampling data from these models. We use group theory for designing efficient architectures and fine-tuning pretrained models. Finally, for sampling from pretrained models, we deploy information-theoretic techniques to obtain high-quality samples. For designing efficient architectures, we provide two novel group equivariant architectures: Equivariant Mesh Attention Networks (EMAN) and Group Representation Networks (G-RepsNet). EMAN is an attention- based architecture for mesh data that is equivariant to various kinds of natural symmetries, including translations, rotations, scaling, node permutations, and gauge transformations. Our pipeline relies on the use of relative tangential features: a simple, effective, equivariance-friendly alternative to raw node positions as inputs. Experiments on standard mesh datasets confirm that our proposed architecture achieves improved performance on these benchmarks and is robust to a wide variety of local/global transformations. G-RepsNet is a scalable lightweight neural network equivariant to arbitrary matrix groups with features represented using tensor polynomials. Existing architectures equivariant to arbitrary matrix groups, in contrast, do not scale well beyond toy datasets. Further, G-RepsNet is also shown to be a universal approximator of functions equivariant to orthogonal groups. Experiments on synthetic datasets, image datasets, and fluid dynamics datasets illustrate its competitive performance to state-of-the-art equivariant models while being computationally inexpensive. For fine-tuning pretrained models, we present equituning, λ-equituning, and multi-equituning. Equituning is a novel finetuning method that transforms (potentially non-equivariant) pretrained models into group equivariant models while incurring a minimum L2 loss between the feature representations of the pretrained and the equivariant models. Large pretrained models can be equituned for different groups to satisfy the needs of various downstream tasks. In λ-equituning, we focus on optimizing the performance of equituning further, and in multi-equituning, we improve the computational efficiency of equituning for large product groups. We test these methods for image classification, compositional generalization in languages, and fairness in natural language generation. For sampling from pretrained large language models (LLMs), we present mirostat, which generates high-quality texts from LLMs by controlling the entropy content of generated texts. It is well known that previous sampling methods such as top-k, top-p, and temperature-based sampling methods often yield texts that have objectionable repetition or incoherence. We find that repetition in generated texts is correlated with low entropy content in them. On the other hand, incoherence in generated text is highly correlated with high entropy content in generated texts. Hence, we provide a sampling algorithm that dynamically controls the entropy content of the generated text. Thus, mirostat helps generate coherent text and avoids repetitions
- …
