Search for: All records

Anomaly/Outlier detection (AD/OD) is often used in controversial applications to detect unusual behavior which is then further investigated or policed. This means an explanation of why something was predicted as an anomaly is desirable not only for individuals but also for the general population and policy-makers. However, existing explainable AI (XAI) methods are not well suited for Explainable Anomaly detection (XAD). In particular, most XAI methods provide instance-level explanations, whereas a model/global-level explanation is desirable for a complete understanding of the definition of normality or abnormality used by an AD algorithm. Further, existing XAI methods try to explain an algorithm’s behavior by finding an explanation of why an instance belongs to a category. However, by definition, anomalies/outliers are chosen because they are different from the normal instances. We propose a new style of model agnostic explanation, called contrastive explanation, that is designed specifically for AD algorithms which use semantic tags to create explanations. It addresses the novel challenge of providing a model-agnostic and global-level explanation by finding contrasts between the outlier group of instances and the normal group. We propose three formulations: (i) Contrastive Explanation, (ii) Strongly Contrastive Explanation, and (iii) Multiple Strong Contrastive Explanations. The last formulation is specifically for the case where a given dataset is believed to have many types of anomalies. For the first two formulations, we show the underlying problem is in the computational class P by presenting linear and polynomial time exact algorithms. We show that the last formulation is computationally intractable, and we use an integer linear program for that version to generate experimental results. We demonstrate our work on several data sets such as the CelebA image data set, the HateXplain language data set, and the COMPAS dataset on fairness. These data sets are chosen as their ground truth explanations are clear or well-known. 

As AI algorithms are deployed extensively, the need to en- sure the fairness of their outputs is critical. Most existing work is on “fairness by design” approaches that incorporate limited tests for fairness into a limited number algorithms. Here, we explore a framework that removes these limitations and can be used with the output of any algorithm that allo- cates instances to one of K categories/classes such as outlier detection (OD), clustering and classification. The framework can encode standard and novel fairness types beyond simple counting, and importantly, it can detect intersectional unfair- ness without being specifically told what to look for. Our ex- perimental results show that both standard and novel types of unfairness exist extensively in the outputs of fair-by-design algorithms and the counter-intuitive observation that they can actually increase intersectional unfairness. 

Deep anomaly detection (AD) is perhaps the most controver- sial of data analytic tasks as it identifies entities that are then specifically targeted for further investigation or exclusion. Also controversial is the application of AI to facial imaging data. This work explores the intersection of these two areas to understand two core questions: ”Who” these algorithms are being unfair to and equally important ”Why”. Recent work has shown that deep AD can be unfair to different groups despite being unsupervised with a recent study showing that for portraits of people: men of color are far more likely to be chosen to be outliers. We study the two main categories of AD algorithms: autoencoder-based and single-class-based which effectively try to compress all the instances with those that can not be easily compressed being deemed to be out- liers. We experimentally verify sources of unfairness such as the under-representation of a group (e.g. people of color are relatively rare), spurious group features (e.g. men are often photographed with hats), and group labeling noise (e.g. race is subjective). We conjecture that lack of compressibility is the main foundation and the others cause it but experimen- tal results show otherwise and we present a natural hierarchy amongst them. 

Societal Impact StatementThe invasive speciesS. alternifloraandP. australisare fast growing coastal wetland plants sequestering large amounts of carbon in the soil and protect coastlines against erosion and storm surges. In this global analysis, we found thatSpartinaandPhragmitesincrease methane but not nitrous oxide emissions, withPhragmiteshaving a lesser effect. The impact of the invasive species on emissions differed greatly among different types of native plant groups, providing valuable information to managers and policymakers during coastal wetland planning and restoration efforts. Further, our estimated net emissions per wetland plant group facilitate regional and national blue carbon estimates. SummaryGlobally,Spartina alternifloraandPhragmites australisare among the most pervasive invasive plants in coastal wetland ecosystems. Both species sequester large amounts of atmospheric carbon dioxide (CO₂) and biogenic carbon in soils but also support production and emission of methane (CH₄). In this study, we investigated the magnitude of their net greenhouse gas (GHG) release from invaded and non‐invaded habitats.We conducted a meta‐analysis of GHG fluxes associated with these two species and related soil carbon content and plant biomass in invaded coastal wetlands.Our results show that both invasive species increase CH₄fluxes compared to uninvaded coastal wetlands, but they do not significantly affect CO₂and N₂O fluxes. The magnitude of emissions fromSpartinaandPhragmitesdiffers among native habitats. GHG fluxes, soil carbon and plant biomass ofSpartina‐invaded habitats were highest compared to uninvaded mudflats and succulent forb‐dominated wetlands, while being lower compared to uninvaded mangroves (except for CH₄).This meta‐analysis highlights the important role of individual plant traits as drivers of change by invasive species on plant‐mediated carbon cycles. 

Knowledge distillation is a simple but powerful way to transfer knowledge between a teacher model to a student model. Existing work suffers from at least one of the following key limitations in terms of direction and scope of transfer which restrict its use: all knowledge is transferred from teacher to student regardless of whether or not that knowledge is useful, the student is the only one learning in this exchange, and typically distillation transfers knowledge only from a single teacher to a single student. We formulate a novel form of knowledge distillation in which many models can act as both students and teachers which we call cooperative distillation. The models cooperate as follows: a model (the student) identifies specific deficiencies in it's performance and searches for another model (the teacher) who encodes learned knowledge into instructional virtual instances via counterfactual instance generation. Because different models may have different strengths and weaknesses, all models can act as either students or teachers (cooperation) when appropriate and only distill knowledge in areas specific to their strengths (focus). Since counterfactuals as a paradigm are not tied to any specific algorithm, we can use this method to distill knowledge between learners of different architectures, algorithms, and even feature spaces. We demonstrate our approach not only outperforms baselines such as transfer learning, self-supervised learning, and multiple knowledge distillation algorithms on several datasets, but it can also be used in settings where the aforementioned techniques cannot. 

Deep learning (DL) is of great interest in psychiatry due its potential yet largely untapped ability to utilize multidimensional datasets (such as fMRI data) to predict clinical outcomes. Typical DL methods, however, have strong assumptions, such as large datasets and underlying model opaqueness, that are suitable for natural image prediction problems but not medical imaging. Here we describe three relatively novel DL approaches that may help accelerate its incorporation into mainstream psychiatry research and ultimately bring it into the clinic as a prognostic tool. We first introduce two methods that can reduce the amount of training data required to develop accurate models. These may prove invaluable for fMRI-based DL given the time and monetary expense required to acquire neuroimaging data. These methods are (1) transfer learning − the ability of deep learners to incorporate knowledge learned from one data source (e.g., fMRI data from one site) and apply it toward learning from a second data source (e.g., data from another site), and (2) data augmentation (via Mixup) − a self-supervised learning technique in which “virtual” instances are created. We then discuss explainable artificial intelligence (XAI), i.e., tools that reveal what features (and in what combinations) deep learners use to make decisions. XAI can be used to solve the “black box” criticism common in DL and reveal mechanisms that ultimately produce clinical outcomes. We expect these techniques to greatly enhance the applicability of DL in psychiatric research and help reveal novel mechanisms and potential pathways for therapeutic intervention in mental illness. 

The key idea of current deep learning methods for dense prediction is to apply a model on a regular patch centered on each pixel to make pixel-wise predictions. These methods are limited in the sense that the patches are determined by network architecture instead of learned from data. In this work, we propose the dense transformer networks, which can learn the shapes and sizes of patches from data. The dense transformer networks employ an encoder-decoder architecture, and a pair of dense transformer modules are inserted into each of the encoder and decoder paths. The novelty of this work is that we provide technical solutions for learning the shapes and sizes of patches from data and efficiently restoring the spatial correspondence required for dense prediction. The proposed dense transformer modules are differentiable, thus the entire network can be trained. We apply the proposed networks on biological image segmentation tasks and show superior performance is achieved in comparison to baseline methods.