An official website of the United States government
We consider the problem of duplicate detection in noisy and incomplete data: Given a large data set in which each record has multiple entries (attributes), detect which distinct records refer to the same real-world entity. This task is complicated by noise (such as misspellings) and missing data, which can lead to records being different, despite referring to the same entity. Our method consists of three main steps: creating a similarity score between records, grouping records together into “unique entities”, and refining the groups. We compare various methods for creating similarity scores between noisy records, considering different combinations of string matching, term frequency-inverse document frequency methods, and n-gram techniques. In particular, we introduce a vectorized soft term frequency-inverse document frequency method, with an optional refinement step. We also discuss two methods to deal with missing data in computing similarity scores. We test our method on the Los Angeles Police Department Field Interview Card data set, the Cora Citation Matching data set, and two sets of restaurant review data. The results show that the methods that use words as the basic units are preferable to those that use 3-grams. Moreover, in some (but certainly not all) parameter ranges soft term frequency-inverse document frequency methods can outperform the standard term frequency-inverse document frequency method. The results also confirm that our method for automatically determining the number of groups typically works well in many cases and allows for accurate results in the absence of a priori knowledge of the number of unique entities in the data set.
more »
« less
Anomaly detection via Gumbel Noise Score Matching
We propose Gumbel Noise Score Matching (GNSM), a novel unsupervised method to detect anomalies in categorical data. GNSM accomplishes this by estimating the scores, i.e., the gradients of log likelihoods w.r.t. inputs, of continuously relaxed categorical distributions. We test our method on a suite of anomaly detection tabular datasets. GNSM achieves a consistently high performance across all experiments. We further demonstrate the flexibility of GNSM by applying it to image data where the model is tasked to detect poor segmentation predictions. Images ranked anomalous by GNSM show clear segmentation failures, with the anomaly scores strongly correlating with segmentation metrics computed on ground-truth. We outline the score matching training objective utilized by GNSM and provide an open-source implementation of our work.
more »
« less
- Award ID(s):
- 2324394
- PAR ID:
- 10628839
- Publisher / Repository:
- Frontiers
- Date Published:
- Journal Name:
- Frontiers in Artificial Intelligence
- Volume:
- 7
- ISSN:
- 2624-8212
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Score matching based diffusion has shown to achieve the state of art results in generation modeling. In the original score matching based diffusion algorithm, the forward equation is a differential equation for which the probability density equation evolves according to a linear partial differential equation, the Fokker-Planck equation. A drawback of this approach is that one needs the data distribution to have a Lipschitz logarithmic gradient. This excludes a large class of data distributions that have a compact support. We present a deterministic diffusion process for which the vector fields are always Lipschitz and hence the score does not explode for probability measures with compact support. This deterministic diffusion process can be seen as a regularization of the porous media equation equation, which enables one to guarantee long term convergence of the forward process to the noise distribution. Though the porous media equation is itself not always guaranteed to have a Lipschitz vector field, it can be used to understand the closeness of the output of the algorithm to the data distribution as a function of the the time horizon and score matching error. This analysis enables us to show that the algorithm has better dependence on the score matching error than approaches based on stochastic diffusions. Using numerical experiments we verify our theoretical results on example one and two dimensional data distributions which are compactly supported. Additionally, we validate the approach on a modified MNIST data set for which the distribution is concentrated on a compact set. In each of the experiments, the approach using deterministic diffusion performs better that the diffusion algorithm with stochastic forward process, when considering the FID scores of the generated samples.more » « less
-
null (Ed.)The ability to detect failures and anomalies are fundamental requirements for building reliable systems for computer vision applications, especially safety-critical applications of semantic segmentation, such as autonomous driving and medical image analysis. In this paper, we systematically study failure and anomaly detection for semantic segmentation and propose a unified framework, consisting of two modules, to address these two related problems. The first module is an image synthesis module, which generates a synthesized image from a segmentation layout map, and the second is a comparison module, which computes the difference between the synthesized image and the input image. We validate our framework on three challenging datasets and improve the state-of-the-arts by large margins, i.e., 6% AUPR-Error on Cityscapes, 7% Pearson correlation on pancreatic tumor segmentation in MSD and 20% AUPR on StreetHazards anomaly segmentatiomore » « less
-
We have been successfully developing Artificial Intelligence (AI) models for automatically classifying fish species using neural networks over the last three years during the “Biology Guided Neural Network” (BGNN) project*1. We continue our efforts in another broader project, “Imageomics: A New Frontier of Biological Information Powered by Knowledge-Guided Machine Learning”*2. One of the main topics in the Imageomics Project is “Morphological Barcoding”. Within the Morphological Barcoding study, we are trying to build a gold standard method to identify species in different taxonomic groups based on their external morphology. This list of characters will contain, but not be limited to, landmarks, quantitative traits such as measurements of distances, areas, angles, proportions, colors, histograms, patterns, shapes, and outlines. The taxonomic groups will be limited by the data available, and we will be using fish as the topic of interest in this preliminary study. In this current study, we have focused on extracting morphological characters that are relying on anatomical features of fish, such as location of the eye, body length, and area of the head. We developed a schematic workflow to describe how we processed the data and extract the information (Fig. 1). We performed our analysis on the segmented images produced by Karpatne Team within the BGNN project (Bart et al. 2021). Segmentation analysis was performed using Artificial Neural Networks - Semantic Segmentation (Long et al. 2015); the list of segments to be detected were given as eye, head, trunk, caudal fin, pectoral fin, dorsal fin, anal fin, pelvic fin for fish. Segmented images, metadata and species lists were given as input to the workflow. During the cleaning and filtering subroutines, a subset of data was created by filtering down to the desired segmented images with corresponding metadata. In the validation step, segmented images were checked by comparing the number of specimens in the original image to the separate bounding-boxed specimen images, noting: violations in the segmentations, counts of segments, comparisons of relative positions of the segments among one another, traces of batch effect; comparisons according to their size and shape and finally based on these validation criteria each segmented image was assigned a score from 1 to 5 similar to Adobe XMP Basic namespace. The landmarks and the traits to be used in the study were extracted from the current literature, while mindful that some of the features may not be extracted successfully computationally. By using the landmark list, landmarks have been extracted by adapting the descriptions from the literature on to the segments, such as picking the left most point on the head as the tip of snout and top left point on the pelvic fin as base of the pelvic fin. These 2D vectors (coordinates), are then fine tuned by adjusting their positions to be on the outline of the fish, since most of the landmarks are located on the outline. Procrustes analysis*3 was performed to scale all of the measurements together and point clouds were generated. These vectors were stored as landmark data. Segment centroids were also treated as landmarks. Extracted landmarks were validated by comparing their relative position among each other, and then if available, compared with their manually captured position. A score was assigned based on these comparisons, similar to the segmentation validation score. Based on the trait list definitions, traits were extracted by measuring the distances between two landmarks, angles between three landmarks, areas between three or more landmarks, areas of the segments, ratios between two distances or areas and between a distance and a square rooted area and then stored as trait data. Finally these values were compared within their own species clusters for errors and whether the values were still within the bounds. Trait scores were calculated from these error calculations similar to segmentation scores aiming selecting good quality scores for further analysis such as Principal Component Analysis. Our work on extraction of features from segmented digital specimen images has shown that the accuracy of the traits such as measurements, areas, and angles depends on the accuracy of the landmarks. Accuracy of the landmarks is highly dependent on segmentation of the parts of the specimen. The landmarks that are located on the outline of the body (combination of head and trunk segments of the fish) are found to be more accurate comparing to the landmarks that represents inner features such as mouth and pectoral fin in some taxonomic groups. However, eye location is almost always accurate, since it is based on the centroid of the eye segment. In the remaining part of this study we will improve the score calculation for segments, images, landmarks and traits and calculate the accuracy of the scores by comparing the statistical results obtained by analysis of the landmark and trait data.more » « less
-
The COVID-19 pandemic has intensified the need for home-based cardiac health monitoring systems. Despite advancements in electrocardiograph (ECG) and phonocardiogram (PCG) wearable sensors, accurate heart sound segmentation algorithms remain understudied. Existing deep learning models, such as convolutional neural networks (CNN) and recurrent neural networks (RNN), struggle to segment noisy signals using only PCG data. We propose a two-step heart sound segmentation algorithm that analyzes synchronized ECG and PCG signals. The first step involves heartbeat detection using a CNN-LSTM-based model on ECG data, and the second step focuses on beat-wise heart sound segmentation with a 1D U-Net that incorporates multi-modal inputs. Our method leverages temporal correlation between ECG and PCG signals to enhance segmentation performance. To tackle the label-hungry issue in AI-supported biomedical studies, we introduce a segment-wise contrastive learning technique for signal segmentation, overcoming the limitations of traditional contrastive learning methods designed for classification tasks. We evaluated our two-step algorithm using the PhysioNet 2016 dataset and a private dataset from Bayland Scientific, obtaining a 96.43 F1 score on the former. Notably, our segment-wise contrastive learning technique demonstrated effective performance with limited labeled data. When trained on just 1% of labeled PhysioNet data, the model pre-trained on the full unlabeled dataset only dropped 2.88 in the F1 score, outperforming the SimCLR method. Overall, our proposed algorithm and learning technique present promise for improving heart sound segmentation and reducing the need for labeled data.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Journal Article:
- https://doi.org/10.3389/frai.2024.1441205
