20 research outputs found
Predicting facebook-users' personality based on status and linguistic features via flexible regression analysis techniques
No full textPredicting facebook-users' personality based on status and linguistic features via flexible regression analysis technique
Three essays on the relationship between land conservation and economic development
No full textLand degradation is a significant cause of biodiversity loss, food insecurity, and persistent poverty. In this dissertation, I explore how land conservation and conversion policies affect economic development and human welfare. I use current and historical contexts and compile primary data sources to answer this question. I use case studies from both developing and developed countries, and from both land conversion and working land conservation policies.
In the first chapter, I study the effects of forest protected areas (PAs) on surrounding households in a developing country. I use Nepal’s recently established PAs as a case study to see the effects on the households who depend on the forest. I find that PAs reduce household wood collection, but there is no evidence that other household consumption is significantly reduced by the strain of reduced access to forest resources nor that PAs rapidly attracted tourism that increased household welfare in these rural villages.
This analysis of the immediate effects of land conservation policies in Nepal cannot shed light on all the effects of conservation policies, as land and soil quality change takes time. Historical events provide a more complete picture. Thus, my second two chapters study historical land conservation policies in the United States (US).
In my second chapter, I explore the persistent impacts on the environment of the earliest farmland conservation policies in the Great Plains. The 1930s Dust Bowl compelled the federal government to undertake large soil conservation policies; I evaluate the effects of those policies over fifty years. Results show that the Voluntary Acreage Reduction program had beneficial long-term effects, increasing areas planted in grassland and decreasing soil erosion in areas that were previously heavily planted in corn and wheat.
Land conservation policies also include creating and nurturing local institutions for management. In my third chapter, I study what factors affected the speed with which local environmental institutions – the Soil Conservation Districts (SCDs) – were created to improve farmland resource management. I use historical documents to create a dataset on exactly when SCDs were established during the period of 1936-1956. A duration analysis of those data finds that SCDs did rise up more rapidly in places hit hardest by crop failure, but institutional change was slower in areas dominated by farms managed by tenants who did not have legal authority to help create SCDs to help preserve their farms.Submission published under a 24 month embargo labeled 'U of I Access', the embargo will last until 2021-08-01The student, Aparna Howlader, accepted the attached license on 2019-07-01 at 13:43.The student, Aparna Howlader, submitted this Dissertation for approval on 2019-07-01 at 14:26.This Dissertation was approved for publication on 2019-07-03 at 16:59.DSpace SAF Submission Ingestion Package generated from Vireo submission #14122 on 2019-11-26 at 13:04:13Made available in DSpace on 2019-11-26T20:49:19Z (GMT). No. of bitstreams: 2
HOWLADER-DISSERTATION-2019.pdf: 48677930 bytes, checksum: fe8bb7b22dcb5fe0facbab6cf2a8b2b5 (MD5)
LICENSE.txt: 4212 bytes, checksum: afa60a3038f33e91bd9002a77e37a64f (MD5)
Previous issue date: 2019-07-03Embargo set by: Seth Robbins for item 112920
Lift date: 2021-11-26T20:49:41Z
Reason: Author requested U of Illinois access only (OA after 2yrs) in Vireo ETD systemU of I Only Restriction Lifted for Item 112920 on 2021-11-27T10:15:33Z
Weighting Pseudo-Labels via High-Activation Feature Index Similarity and Object Detection for Semi-Supervised Segmentation
No full textSemi-supervised semantic segmentation methods leverage unlabeled data by pseudo-labeling them. Thus the success of these methods hinges on the reliablility of the pseudo-labels. Existing methods mostly choose high-confidence pixels in an effort to avoid erroneous pseudo-labels. However, high confidence does not guarantee correct pseudo-labels especially in the initial training iterations. In this paper, we propose a novel approach to reliably learn from pseudo-labels. First, we unify the predictions from a trained object detector and a semantic segmentation model to identify reliable pseudo-label pixels. Second, we assign different learning weights to pseudo-labeled pixels to avoid noisy training signals. To determine these weights, we first use the reliable pseudo-label pixels identified from the first step and labeled pixels to construct a prototype for each class. Then, the per-pixel weight is the structural similarity between the pixel and the prototype measured via rank-statistics similarity. This metric is robust to noise, making it better suited for comparing features from unlabeled images, particularly in the initial training phases where wrong pseudo labels are prone to occur. We show that our method can be easily integrated into four semi-supervised semantic segmentation frameworks, and improves them in both Cityscapes and Pascal VOC datasets.to be published in ECCV2
Weighting Pseudo-labels via High-Activation Feature Index Similarity and Object Detection for Semi-supervised Segmentation
No full textSemi-supervised semantic segmentation methods leverage unlabeled data by pseudo-labeling them. Thus the success of these methods hinges on the reliability of the pseudo-labels. Existing methods mostly choose high-confidence pixels in an effort to avoid erroneous pseudo-labels. However, high confidence does not guarantee correct pseudo-labels especially in the initial training iterations. In this paper, we propose a novel approach to reliably learn from pseudo-labels. First, we unify the predictions from a trained object detector and a semantic segmentation model to identify reliable pseudo-label pixels. Second, we assign different learning weights to pseudo-labeled pixels to avoid noisy training signals. To determine these weights, we first use the reliable pseudo-label pixels identified from the first step and labeled pixels to construct a prototype for each class. Then, the per-pixel weight is the similarity score between the pixel and the prototype measured via rank-statistics. This metric is robust to noise, making it better suited for comparing features from unlabeled images, particularly in the initial training phases where wrong pseudo labels are prone to occur. We show that our method can be easily integrated into four semi-supervised semantic segmentation frameworks, and improves them in both Cityscapes and Pascal VOC datasets. Code is available at https://github.com/cvlab-stonybrook/Weighting-Pseudo-Labels.CVLA
Grad-CAM++: Generalized Gradient-Based Visual Explanations for Deep Convolutional Networks
No full textOver the last decade, Convolutional Neural Network (CNN) models have been highly successful in solving complex vision based problems. However, deep models are perceived as "black box" methods considering the lack of understanding of their internal functioning. There has been a significant recent interest to develop explainable deep learning models, and this paper is an effort in this direction. Building on a recently proposed method called Grad-CAM, we propose Grad-CAM++ to provide better visual explanations of CNN model predictions (when compared to Grad-CAM), in terms of better localization of objects as well as explaining occurrences of multiple objects of a class in a single image. We provide a mathematical explanation for the proposed method, Grad-CAM++, which uses a weighted combination of the positive partial derivatives of the last convolutional layer feature maps with respect to a specific class score as weights to generate a visual explanation for the class label under consideration. Our extensive experiments and evaluations, both subjective and objective, on standard datasets showed that Grad-CAM++ indeed provides better visual explanations for a given CNN architecture when compared to Grad-CAM
Beyond Pixels: Semi-Supervised Semantic Segmentation with a Multi-scale Patch-based Multi-Label Classifier
No full textIncorporating pixel contextual information is critical for accurate segmentation. In this paper, we show that an effective way to incorporate contextual information is through a patch-based classifier. This patch classifier is trained to identify classes present within an image region, which facilitates the elimination of distractors and enhances the classification of small object segments. Specifically, we introduce Multi-scale Patch-based Multi-label Classifier (MPMC), a novel plug-in module designed for existing semi-supervised segmentation (SSS) frameworks. MPMC offers patch-level supervision, enabling the discrimination of pixel regions of different classes within a patch. Furthermore, MPMC learns an adaptive pseudo-label weight, using patch-level classification to alleviate the impact of the teacher\u27s noisy pseudo-label supervision the student. This lightweight module can be integrated into any SSS framework, significantly enhancing their performance. We demonstrate the efficacy of our proposed MPMC by integrating it into four SSS methodologies and improving them across two natural image and one medical segmentation dataset, notably improving the segmentation results of the baselines across all the three datasets.to be published in ECCV2
RESOURCE ALLOCATION IN 802.11AX NETWORKS
No full textMethods of selection of Voice over Internet Protocol (VoIP), video and other users to meet quality of service (QoS) goals and optimize overall performance in 802.11ax networks are provided. These methods allow policy based decisions such as controlling the number of video, VoIP or other users or sub-channel sizes for video (or other) users or deciding data rate (or associated modulation and coding scheme) for each user in each scheduling interval (SI), and allow dynamic decisions for the value of the SI
The Rail-bridge Interaction – Recent Advances with ERS Fastening System for Steel Bridges
No full textAbstractThe requirements on the existing rail infrastructure, including bridges, need for higher speed as well as to accommodate the extremely growing traffic demand has brought major changes in different solution to railway track systems in last few decades. Since the beginning of the direct fastening system for railway bridges, probably Embedded Rail System (ERS) is one of the most interesting one. Especially, the low maintenance requirements together with the capability to refurbishment of existing bridges and low noise emission has given new possibilities to this system, contributing towards environmental and economic sustainability. The purpose of this paper is to analyze and describe the response of ERS system under different vertical and horizontal load, based on a small scale laboratory test. Second part of the paper describes the numerical analysis of the application of the ERS system on the Starý most in Bratislava, where the test results were used
