Most post-training backdoor detection methods rely on attacked models exhibiting extreme outlier detection statistics for the target class of an attack, compared to non-target classes. However, these approaches may fail: (1) when some (non-target) classes are easily discriminable from all others, in which case they may naturally achieve extreme detection statistics (e.g., decision confidence); and (2) when the backdoor is subtle, i.e., with its features weak relative to intrinsic class-discriminative features. A key observation is that the backdoor target class has contributions to its detection statistic from both the backdoor trigger and from its intrinsic features, whereas non-target classes only have contributions from their intrinsic features. To achieve more sensitive detectors, we thus propose to suppress intrinsic features while optimizing the detection statistic for a given class. For non-target classes, such suppression will drastically reduce the achievable statistic, whereas for the target class the (significant) contribution from the backdoor trigger remains. In practice, we formulate a constrained optimization problem, leveraging a small set of clean examples from a given class, and optimizing the detection statistic while orthogonalizing with respect to the class's intrinsic features. We dub this plug-and-play approach Class Subspace Orthogonalization (CSO) and assess it against challenging mixed-label and adaptive attacks.

The manifold hypothesis is a core mechanism behind the success of deep learning, so understanding the intrinsic manifold structure of image data is central to studying how neural networks learn from the data. Intrinsic dataset manifolds and their relationship to learning difficulty have recently begun to be studied for the common domain of natural images, but little such research has been attempted for radiological images. We address this here. First, we compare the intrinsic manifold dimensionality of radiological and natural images. We also investigate the relationship between intrinsic dimensionality and generalization ability over a wide range of datasets. Our analysis shows that natural image datasets generally have a higher number of intrinsic dimensions than radiological images. However, the relationship between generalization ability and intrinsic dimensionality is much stronger for medical images, which could be explained as radiological images having intrinsic features that are more difficult to learn. These results give a more principled underpinning for the intuition that radiological images can be more challenging to apply deep learning to than natural image datasets common to machine learning research. We believe rather than directly applying models developed for natural images to the radiological imaging domain, more care should be taken to developing architectures and algorithms that are more tailored to the specific characteristics of this domain. The research shown in our paper, demonstrating these characteristics and the differences from natural images, is an important first step in this direction.

Insurance fraud occurs when policyholders file claims that are exaggerated or based on intentional damages. This contribution develops a fraud detection strategy by extracting insightful information from the social network of a claim. First, we construct a network by linking claims with all their involved parties, including the policyholders, brokers, experts, and garages. Next, we establish fraud as a social phenomenon in the network and use the BiRank algorithm with a fraud specific query vector to compute a fraud score for each claim. From the network, we extract features related to the fraud scores as well as the claims' neighborhood structure. Finally, we combine these network features with the claim-specific features and build a supervised model with fraud in motor insurance as the target variable. Although we build a model for only motor insurance, the network includes claims from all available lines of business. Our results show that models with features derived from the network perform well when detecting fraud and even outperform the models using only the classical claim-specific features. Combining network and claim-specific features further improves the performance of supervised learning models to detect fraud. The resulting model flags highly suspicions claims that need to be further investigated. Our approach provides a guided and intelligent selection of claims and contributes to a more effective fraud investigation process.

Most physical or social phenomena can be represented by ontologies where the constituent entities are interacting in various ways with each other and with their environment. Furthermore, those entities are likely heterogeneous and attributed with features that evolve dynamically in time as a response to their successive interactions. In order to apply machine learning on such entities, e.g., for classification purposes, one therefore needs to integrate the interactions into the feature engineering in a systematic way. This proposal shows how, to this end, the current state of graph machine learning remains inadequate and needs to be be augmented with a comprehensive feature engineering paradigm in space and time.

Meaning may arise from an element's role or interactions within a larger system. For example, hitting nails is more central to people's concept of a hammer than its particular material composition or other intrinsic features. Likewise, the importance of a web page may result from its links with other pages rather than solely from its content. One example of meaning arising from extrinsic relationships are approaches that extract the meaning of word concepts from co-occurrence patterns in large, text corpora. The success of these methods suggest that human activity patterns may reveal conceptual organization. However, texts do not directly reflect human activity, but instead serve a communicative function and are usually highly curated or edited to suit an audience. Here, we apply methods devised for text to a data source that directly reflects thousands of individuals' activity patterns, namely supermarket purchases. Using product co-occurrence data from nearly 1.3m shopping baskets, we trained a topic model to learn 25 high-level concepts (or "topics"). These topics were found to be comprehensible and coherent by both retail experts and consumers. Topics ranged from specific (e.g., ingredients for a stir-fry) to general (e.g., cooking from scratch). Topics tended to be goal-directed and situational, consistent with the notion that human conceptual knowledge is tailored to support action. Individual differences in the topics sampled predicted basic demographic characteristics. These results suggest that human activity patterns reveal conceptual organization and may give rise to it.

Keyword extraction attracts much attention for its significant role in various natural language processing tasks. While some existing methods for keyword extraction have considered using single type of semantic relatedness between words or inherent attributes of words, almost all of them ignore two important issues: 1) how to fuse multiple types of semantic relations between words into a uniform semantic measurement and automatically learn the weights of the edges between the words in the word graph of each document, and 2) how to integrate the relations between words and words' intrinsic features into a unified model. In this work, we tackle the two issues based on the supervised random walk model. We propose a supervised ranking based method for keyword extraction, which is called SEAFARER. It can not only automatically learn the weights of the edges in the unified graph of each document which includes multiple semantic relations but also combine the merits of semantic relations of edges and intrinsic attributes of nodes together. We conducted extensive experimental study on an established benchmark and the experimental results demonstrate that SEAFARER outperforms the state-of-the-art supervised and unsupervised methods.