Search

We construct a Poincar\'e complex whose periodic total surgery obstruction vanishes but whose Spivak normal fibration does not admit a reduction to a stable euclidean bundle. This contradicts the conjunction of two claims in the literature: Namely, on the one hand that a Poincar\'e complex with vanishing periodic total surgery obstruction is homotopy equivalent to a homology manifold, which appears in work of Bryant--Ferry--Mio--Weinberger, and on the other that the Spivak normal fibration of a homology manifold always admits a reduction to a stable euclidean bundle, which appears in work of Ferry--Pedersen.

Deep learning has become the de facto method for medical image segmentation, with 3D segmentation models excelling in capturing complex 3D structures and 2D models offering high computational efficiency. However, segmenting 2.5D images, which have high in-plane but low through-plane resolution, is a relatively unexplored challenge. While applying 2D models to individual slices of a 2.5D image is feasible, it fails to capture the spatial relationships between slices. On the other hand, 3D models face challenges such as resolution inconsistencies in 2.5D images, along with computational complexity and susceptibility to overfitting when trained with limited data. In this context, 2.5D models, which capture inter-slice correlations using only 2D neural networks, emerge as a promising solution due to their reduced computational demand and simplicity in implementation. In this paper, we introduce CSA-Net, a flexible 2.5D segmentation model capable of processing 2.5D images with an arbitrary number of slices through an innovative Cross-Slice Attention (CSA) module. This module uses the cross-slice attention mechanism to effectively capture 3D spatial information by learning long-range dependencies between the center slice (for segmentation) and its neighboring slices. Moreover, CSA-Net utilizes the self-attention mechanism to understand correlations among pixels within the center slice. We evaluated CSA-Net on three 2.5D segmentation tasks: (1) multi-class brain MRI segmentation, (2) binary prostate MRI segmentation, and (3) multi-class prostate MRI segmentation. CSA-Net outperformed leading 2D and 2.5D segmentation methods across all three tasks, demonstrating its efficacy and superiority. Our code is publicly available at https://github.com/mirthAI/CSA-Net.

We show that in codimension at least 3, spaces of locally flat topological embeddings of manifolds are correctly modelled by derived spaces of maps between their configuration categories (under mild smoothability conditions). That general claim was reduced in an earlier paper to the special cases where the manifolds in question are euclidean spaces. We deal with these special cases by comparing to other special cases where the manifolds have the form "torus" and "torus times euclidean space" respectively, and by setting up a torus trick for configuration categories., Comment: 52 pages. Changes in v3: some minor changes to sections 4 and appendix A and bigger changes to sections 6 and 7

The configuration category of a manifold is a topological category which we view as a Segal space, via the nerve construction. Our main result is that the unordered configuration category, suitably truncated, admits a finite presentation as a complete Segal space if the manifold in question is the interior of a compact manifold., Comment: 15 pages

We investigate the relationship between the configuration category of a manifold and the configuration category of a covering space of that manifold., Comment: 18 pages

In two parts, we present a bigness criterion for the cotangent bundle of resolutions of orbifold surfaces of general type. As a corollary, we obtain the \textit{canonical model singularities} (CMS) criterion that can be applied to determine when a birational class of surfaces has smooth representatives with big cotangent bundle and compare it with other known criteria. We then apply this criterion to the problem of finding the minimal degrees $d$ for which the deformation equivalence class of a smooth hypersurface of degree d in $\PP^3$ has a representative with big cotangent bundle; applications to the minimal resolutions of cyclic covers of $\PP^2$ branched along line arrangements in general position are also obtained. The CMS criterion involves invariants of canonical singularities whose values were unknown. In this part of the work, we describe a method of finding these invariants and obtain formulas for $A_n$-singularities. We also use our approach to derive several extension results for symmetric differentials on the complement of the exceptional locus $E$ of the minimal resolution of an $A_n$-singularity; namely, we characterize the precise extent to which the poles along $E$ of symmetric differentials on the complement are milder than logarithmic poles., Comment: 40 pages

Community colleges provide postsecondary education to millions of students in the United States each year, but their graduation rates are low. Many community colleges have implemented interventions to help students persist in college and earn degrees. MDRC has studied many such interventions; several of them improved students' academic outcomes, but the effects varied. This report synthesizes results from MDRC's studies. The data include findings from 30 studies of 39 interventions at 45 colleges. The interventions included different components designed to help students succeed: enhanced advising; enhanced tutoring; financial support; instructional reform; learning communities (which enroll small groups of students together in two or more courses); promoting full-time enrollment in fall and spring, summer enrollment, or both; and success courses (designed to help new students navigate college and build relevant skills). MDRC conducted a quantitative synthesis of the studies' results to examine the relationship between various intervention features and effects on selected student outcomes. This report draws from the quantitative synthesis and also from detailed information on how the interventions were implemented. The research also points to some areas for further exploration. Those areas include how to support part-time community college students effectively, the implementation and effects of programs targeting students in rural areas and in small colleges, and the potential role of community college interventions in reducing racial inequality in students' academic outcomes.

What works to help community college students progress academically? This brief synthesizes 20 years of rigorous research by MDRC, presenting new evidence about key attributes of community college interventions that are positively related to larger impacts on students' academic progress. Findings: Findings are based on a synthesis of evidence from 30 randomized controlled trials of 39 postsecondary interventions involving 60,000 students. The results of this research consistently indicate that the impacts of community college interventions increase with: (1) The comprehensiveness of the intervention, as measured by its number of components; and (2) The promotion of full-time enrollment (during fall and spring) and summer enrollment. Less consistent, but still promising evidence suggests the impacts of community college interventions also increase with the extent that they increase: (1) Advising use among students; (2) Tutoring use among students; and (3) Financial support for students (though the evidence is least consistent for this component). Consequently, the preceding five intervention features seem like a reasonable, evidence-based place to start when developing community college policy, designing a new community college intervention, or enhancing an existing community college intervention.

Recent decades have seen the rise of large-scale Deep Neural Networks (DNNs) to achieve human-competitive performance in a variety of artificial intelligence tasks. Often consisting of hundreds of millions, if not hundreds of billion parameters, these DNNs are too large to be deployed to, or efficiently run on resource-constrained devices such as mobile phones or IoT microcontrollers. Systems relying on large-scale DNNs thus have to call the corresponding model over the network, leading to substantial costs for hosting and running the large-scale remote model, costs which are often charged on a per-use basis. In this paper, we propose BiSupervised, a novel architecture, where, before relying on a large remote DNN, a system attempts to make a prediction on a small-scale local model. A DNN supervisor monitors said prediction process and identifies easy inputs for which the local prediction can be trusted. For these inputs, the remote model does not have to be invoked, thus saving costs, while only marginally impacting the overall system accuracy. Our architecture furthermore foresees a second supervisor to monitor the remote predictions and identify inputs for which not even these can be trusted, allowing to raise an exception or run a fallback strategy instead. We evaluate the cost savings, and the ability to detect incorrectly predicted inputs on four diverse case studies: IMDB movie review sentiment classification, Github issue triaging, Imagenet image classification, and SQuADv2 free-text question answering, Comment: A corresponding registered Report (i.e., paper without results) was accepted at ACM TOSEM. This preprint includes the (not yet peer-reviewed) results

Deep Neural Networks (DNN) are increasingly used as components of larger software systems that need to process complex data, such as images, written texts, audio/video signals. DNN predictions cannot be assumed to be always correct for several reasons, among which the huge input space that is dealt with, the ambiguity of some inputs data, as well as the intrinsic properties of learning algorithms, which can provide only statistical warranties. Hence, developers have to cope with some residual error probability. An architectural pattern commonly adopted to manage failure-prone components is the supervisor, an additional component that can estimate the reliability of the predictions made by untrusted (e.g., DNN) components and can activate an automated healing procedure when these are likely to fail, ensuring that the Deep Learning based System (DLS) does not cause damages, despite its main functionality being suspended. In this paper, we consider DLS that implement a supervisor by means of uncertainty estimation. After overviewing the main approaches to uncertainty estimation and discussing their pros and cons, we motivate the need for a specific empirical assessment method that can deal with the experimental setting in which supervisors are used, where the accuracy of the DNN matters only as long as the supervisor lets the DLS continue to operate. Then we present a large empirical study conducted to compare the alternative approaches to uncertainty estimation. We distilled a set of guidelines for developers that are useful to incorporate a supervisor based on uncertainty monitoring into a DLS., Comment: Accepted for publication at the Journal of Software: Testing, Verification and Reliability. arXiv admin note: substantial text overlap with arXiv:2102.00902

On complex problems, state of the art prediction accuracy of Deep Neural Networks (DNN) can be achieved using very large-scale models, consisting of billions of parameters. Such models can only be run on dedicated servers, typically provided by a 3rd party service, which leads to a substantial monetary cost for every prediction. We propose a new software architecture for client-side applications, where a small local DNN is used alongside a remote large-scale model, aiming to make easy predictions locally at negligible monetary cost, while still leveraging the benefits of a large model for challenging inputs. In a proof of concept we reduce prediction cost by up to 50% without negatively impacting system accuracy., Comment: Research Abstract. Contact me for a pre-print of the full paper (currently not yet published)

Deep Neural Networks (DNNs) are becoming a crucial component of modern software systems, but they are prone to fail under conditions that are different from the ones observed during training (out-of-distribution inputs) or on inputs that are truly ambiguous, i.e., inputs that admit multiple classes with nonzero probability in their labels. Recent work proposed DNN supervisors to detect high-uncertainty inputs before their possible misclassification leads to any harm. To test and compare the capabilities of DNN supervisors, researchers proposed test generation techniques, to focus the testing effort on high-uncertainty inputs that should be recognized as anomalous by supervisors. However, existing test generators aim to produce out-of-distribution inputs. No existing model- and supervisor independent technique targets the generation of truly ambiguous test inputs, i.e., inputs that admit multiple classes according to expert human judgment. In this paper, we propose a novel way to generate ambiguous inputs to test DNN supervisors and used it to empirically compare several existing supervisor techniques. In particular, we propose AmbiGuess to generate ambiguous samples for image classification problems. AmbiGuess is based on gradient-guided sampling in the latent space of a regularized adversarial autoencoder. Moreover, we conducted what is -- to the best of our knowledge -- the most extensive comparative study of DNN supervisors, considering their capabilities to detect 4 distinct types of high-uncertainty inputs, including truly ambiguous ones. We find that the tested supervisors' capabilities are complementary: Those best suited to detect true ambiguity perform worse on invalid, out-of-distribution and adversarial inputs and vice-versa., Comment: Accepted for publication at Springers "Empirical Software Engineering" (EMSE)

Some education programs' early positive effects disappear over time. Other programs have unanticipated positive long-term effects. Foundations warn of the dangers of putting too much weight on in-program effects, which, they say, often fade after a program ends. This Issue Focus tackles the topic of post-program effects in postsecondary education. Are in-program effects--that is, the effects observed while the program was active--maintained once the program ends? Do they grow and improve? Or do they fade out? This investigation capitalizes on two decades of rigorous program evaluations conducted by MDRC, including approximately 25 postsecondary programs, to consider those questions. The Higher Education Randomized Controlled Trials (THE RCT) draw from a database of student-level data from 31 MDRC projects, sampling 67,400 stu­dents. The results of THE RCT are striking: During the year after these programs ended, effects on academic progress were consistently maintained. There is no evidence of discernible fade-out, which provides encouraging information about the lasting value of many postsecondary programs and the value of evidence collected from them. There was also no evidence of post-program growth, perhaps prompting the need to develop programs that equip students for success over the long term.

Most community college students are referred to developmental education courses to build basic skills. These students often struggle in these courses and college more broadly. CUNY Start is a prematriculation program for students assessed as having significant remedial needs. CUNY Start students delay matriculation for one semester and receive time-intensive instruction in math, reading, and writing with a prescribed pedagogy delivered by trained teachers. The program aims to help students complete remediation and prepare for college-level courses. This article describes the results of an experiment at four community colleges (n [is approximately equal to] 3,800). We estimate that over three years, including one semester that students spent in the program and two-and-a-half years after the program was complete, CUNY Start substantially increased college readiness, slightly increased credit accumulation, and modestly increased graduation rates (by increasing participation in CUNY's highly effective ASAP).

Test Input Prioritizers (TIP) for Deep Neural Networks (DNN) are an important technique to handle the typically very large test datasets efficiently, saving computation and labeling costs. This is particularly true for large-scale, deployed systems, where inputs observed in production are recorded to serve as potential test or training data for the next versions of the system. Feng et. al. propose DeepGini, a very fast and simple TIP, and show that it outperforms more elaborate techniques such as neuron- and surprise coverage. In a large-scale study (4 case studies, 8 test datasets, 32'200 trained models) we verify their findings. However, we also find that other comparable or even simpler baselines from the field of uncertainty quantification, such as the predicted softmax likelihood or the entropy of the predicted softmax likelihoods perform equally well as DeepGini., Comment: Accepted at ISSTA 2022

This paper presents new estimates of the effects of the City University of New York (CUNY) Accelerated Study in Associate Programs (ASAP) model, evaluated using a randomized controlled trial first in New York and later through a replication in Ohio. It describes longer-term effects of CUNY ASAP in New York, showing that the program's effects on associate's degree receipt persisted through eight years and likely represent a permanent increase in degree receipt. The paper also offers an analysis from the pooled study samples in New York and Ohio. The findings indicate that the program has consistent effects on degree receipt across different states but also for somewhat different levels of service contrast, such as the number of additional advising visits.

This report presents findings from Encouraging Additional Summer Enrollment [EASE], which used behavioral insights in two informational campaigns, with and without tuition assistance, to encourage community college students to take summer classes. Both interventions increased enrollment and had a modest impact on credits earned and positive return on investment for colleges. [The Encouraging Additional Summer Enrollment (EASE) project is funded by Ascendium Education Group. This report was written with Xavier Alemañy.]

Surprise Adequacy (SA) is one of the emerging and most promising adequacy criteria for Deep Learning (DL) testing. As an adequacy criterion, it has been used to assess the strength of DL test suites. In addition, it has also been used to find inputs to a Deep Neural Network (DNN) which were not sufficiently represented in the training data, or to select samples for DNN retraining. However, computation of the SA metric for a test suite can be prohibitively expensive, as it involves a quadratic number of distance calculations. Hence, we developed and released a performance-optimized, but functionally equivalent, implementation of SA, reducing the evaluation time by up to 97\%. We also propose refined variants of the SA omputation algorithm, aiming to further increase the evaluation speed. We then performed an empirical study on MNIST, focused on the out-of-distribution detection capabilities of SA, which allowed us to reproduce parts of the results presented when SA was first released. The experiments show that our refined variants are substantially faster than plain SA, while producing comparable outcomes. Our experimental results exposed also an overlooked issue of SA: it can be highly sensitive to the non-determinism associated with the DNN training procedure., Comment: Accepted at DeepTest 2021 (ICSE Workshop)

Modern software systems rely on Deep Neural Networks (DNN) when processing complex, unstructured inputs, such as images, videos, natural language texts or audio signals. Provided the intractably large size of such input spaces, the intrinsic limitations of learning algorithms, and the ambiguity about the expected predictions for some of the inputs, not only there is no guarantee that DNN's predictions are always correct, but rather developers must safely assume a low, though not negligible, error probability. A fail-safe Deep Learning based System (DLS) is one equipped to handle DNN faults by means of a supervisor, capable of recognizing predictions that should not be trusted and that should activate a healing procedure bringing the DLS to a safe state. In this paper, we propose an approach to use DNN uncertainty estimators to implement such a supervisor. We first discuss the advantages and disadvantages of existing approaches to measure uncertainty for DNNs and propose novel metrics for the empirical assessment of the supervisor that rely on such approaches. We then describe our publicly available tool UNCERTAINTY-WIZARD, which allows transparent estimation of uncertainty for regular tf.keras DNNs. Lastly, we discuss a large-scale study conducted on four different subjects to empirically validate the approach, reporting the lessons-learned as guidance for software engineers who intend to monitor uncertainty for fail-safe execution of DLS., Comment: Accepted at IEEE International Conference on Software Testing, Verification and Validation 2021

Uncertainty and confidence have been shown to be useful metrics in a wide variety of techniques proposed for deep learning testing, including test data selection and system supervision.We present uncertainty-wizard, a tool that allows to quantify such uncertainty and confidence in artificial neural networks. It is built on top of the industry-leading tf.keras deep learning API and it provides a near-transparent and easy to understand interface. At the same time, it includes major performance optimizations that we benchmarked on two different machines and different configurations., Comment: Accepted for publication at the IEEE International Conference on Software Testing, Verification and Validation 2021