{"arxiv_id":"2112.07662","incoming":{"papers":0,"sentences":0,"groups":[],"unlabelled_sentences":0},"outgoing":{"papers":16,"sentences":34,"groups":[{"kind":"extends","label":"Builds on","papers":2,"sentences":2,"quotes":[{"arxiv_id":"1906.12340","title":"Using Self-Supervised Learning Can Improve Model Robustness and Uncertainty","url":"/paper/1906.12340","section":"Feature-Adaptation for Labeled Multi-Class Anomaly Detection","sentence":"Here, we extend the work of Hendrycks et al. [ref] and Fort et al. [ref] which suggested initializing the classifier $C$ with weights that are pre-trained on a large-scale datasets.","intent_p":0.92,"contests_p":null},{"arxiv_id":"2106.03004","title":"Exploring the Limits of Out-of-Distribution Detection","url":"/paper/2106.03004","section":"Feature-Adaptation for Labeled Multi-Class Anomaly Detection","sentence":"Here, we extend the work of Hendrycks et al. [ref] and Fort et al. [ref] which suggested initializing the classifier $C$ with weights that are pre-trained on a large-scale datasets.","intent_p":0.92,"contests_p":null}]},{"kind":"method_or_tool","label":"Uses the method or tool of","papers":3,"sentences":4,"quotes":[{"arxiv_id":"2106.03004","title":"Exploring the Limits of Out-of-Distribution Detection","url":"/paper/2106.03004","section":"Finetuning Pre-trained Features with Pseudo Labels","sentence":"Recent methods suggested a per-class Mahalanobis approach [ref] deeming a sample normal if it lies within a small Mahalanobis distance to any of the cluster centers.","intent_p":0.8,"contests_p":null},{"arxiv_id":"2106.03844","title":"Mean-Shifted Contrastive Loss for Anomaly Detection","url":"/paper/2106.03844","section":"Multi-Class Anomaly Detection With Pre-trained Features","sentence":"We follow MeanShifted [ref] by using the cosine similarity distance instead of $l_2$ distance.","intent_p":1.0,"contests_p":null},{"arxiv_id":"2005.02357","title":"Sub-Image Anomaly Detection with Deep Pyramid Correspondences","url":"/paper/2005.02357","section":"Discussion","sentence":"Following previous works [ref] , all our main results are reported using $k$NN with $k=1$.","intent_p":0.91,"contests_p":null}]},{"kind":"data_used","label":"Uses data or values from","papers":1,"sentences":1,"quotes":[{"arxiv_id":"1506.03365","title":"LSUN: Construction of a Large-scale Image Dataset using Deep Learning with Humans in the Loop","url":"/paper/1506.03365","section":"Results","sentence":"We evaluate it against a variety of dataset supplied in a similar resolution, namely: CIFAR-100 [ref] , SVHN [ref] , LSUN [ref] .","intent_p":0.88,"contests_p":null}]},{"kind":"comparison","label":"Compares with","papers":2,"sentences":2,"quotes":[{"arxiv_id":"2106.03844","title":"Mean-Shifted Contrastive Loss for Anomaly Detection","url":"/paper/2106.03844","section":"Introduction","sentence":"We find that our method outperforms state-of-the-art self-supervised [ref] and pre-trained [ref] anomaly detection methods.","intent_p":1.0,"contests_p":null},{"arxiv_id":"2106.03004","title":"Exploring the Limits of Out-of-Distribution Detection","url":"/paper/2106.03004","section":"Discussion","sentence":"For example, in the CIFAR-10 vs. CIFAR-100 multi-class anomaly detection task, our method achieves 96.7% ROCAUC, while a similar method by Fort et al. [ref] achieves $98.4$% ROCAUC using the ground truth labels.","intent_p":0.99,"contests_p":null}]},{"kind":"supports_this_paper","label":"Cites as support","papers":1,"sentences":1,"quotes":[{"arxiv_id":"2111.09832","title":"Merging Models with Fisher-Weighted Averaging","url":"/paper/2111.09832","section":"Discussion","sentence":"This phenomenon has also been noticed by previous works [ref] .","intent_p":0.97,"contests_p":null}]},{"kind":"background","label":"Cites as background","papers":4,"sentences":4,"quotes":[{"arxiv_id":"1906.02845","title":"Likelihood Ratios for Out-of-Distribution Detection","url":"/paper/1906.02845","section":"Multi-Class Anomaly Detection Without Pretraining","sentence":"Different methods suggested modification to this score to account for the dataset statistics (Likelihood Ratio [ref] ) or its complexity (Input Complexity [ref] ).","intent_p":0.82,"contests_p":null},{"arxiv_id":"1909.11480","title":"Input complexity and out-of-distribution detection with likelihood-based generative models","url":"/paper/1909.11480","section":"Multi-Class Anomaly Detection Without Pretraining","sentence":"Different methods suggested modification to this score to account for the dataset statistics (Likelihood Ratio [ref] ) or its complexity (Input Complexity [ref] ).","intent_p":0.82,"contests_p":null},{"arxiv_id":"2205.09310","title":"Mitigating Neural Network Overconfidence with Logit Normalization","url":"/paper/2205.09310","section":"Feature-Adaptation for Labeled Multi-Class Anomaly Detection","sentence":"Many techniques have been proposed to improve the obtained OOD detection [ref] .","intent_p":0.83,"contests_p":null},{"arxiv_id":"2108.04116","title":"Transfer Learning Gaussian Anomaly Detection by Fine-tuning Representations","url":"/paper/2108.04116","section":"Discussion","sentence":"Previous works used a variety of scoring criterion for multi-class anomaly detection [ref] .","intent_p":1.0,"contests_p":null}]}],"unlabelled_sentences":20},"claims":[],"thresholds":{"intent_p":0.8,"supports_p":0.95,"contests_p":0.7},"method":"Citing sentences come from each paper's LaTeX source; every \\cite marks the sentence and the works it cites. Pith's classifier typed each sentence. A label is shown at intent probability 0.8 or more (0.95 for support), a contest at 0.7 or more. Measured accuracy against two blind model families is at https://pith.science/accuracy. The quoted sentences are the citing authors' own.","held_up":null,"_links":{"html":"https://pith.science/paper/2112.07662"}}