| 1. |
- Berg, Amanda, 1988-, et al.
(författare)
-
Unsupervised Adversarial Learning of Anomaly Detection in the Wild
- 2020
-
Ingår i: Proceedings of the 24th European Conference on Artificial Intelligence (ECAI). - Amsterdam : IOS Press. - 9781643681009 - 9781643681016 ; , s. 1002-1008
-
Konferensbidrag (refereegranskat)abstract
- Unsupervised learning of anomaly detection in high-dimensional data, such as images, is a challenging problem recently subject to intense research. Through careful modelling of the data distribution of normal samples, it is possible to detect deviant samples, so called anomalies. Generative Adversarial Networks (GANs) can model the highly complex, high-dimensional data distribution of normal image samples, and have shown to be a suitable approach to the problem. Previously published GAN-based anomaly detection methods often assume that anomaly-free data is available for training. However, this assumption is not valid in most real-life scenarios, a.k.a. in the wild. In this work, we evaluate the effects of anomaly contaminations in the training data on state-of-the-art GAN-based anomaly detection methods. As expected, detection performance deteriorates. To address this performance drop, we propose to add an additional encoder network already at training time and show that joint generator-encoder training stratifies the latent space, mitigating the problem with contaminated data. We show experimentally that the norm of a query image in this stratified latent space becomes a highly significant cue to discriminate anomalies from normal data. The proposed method achieves state-of-the-art performance on CIFAR-10 as well as on a large, previously untested dataset with cell images.
|
|
| 2. |
- Bhatt, Mehul, Professor, 1980-, et al.
(författare)
-
Cognitive Vision and Perception : Deep Semantics Integrating AI and Vision for Reasoning about Space, Motion, and Interaction
- 2020
-
Ingår i: ECAI 2020. - : IOS Press. - 9781643681009 - 9781643681016 ; , s. 2881-2882
-
Konferensbidrag (refereegranskat)abstract
- Semantic interpretation of dynamic visuospatial imagery calls for a general and systematic integration of methods in knowledge representation and computer vision. Towards this, we highlight research articulating & developing deep semantics, characterised by the existence of declarative models –e.g., pertaining space and motion– and corresponding formalisation and reasoning methods sup- porting capabilities such as semantic question-answering, relational visuospatial learning, and (non-monotonic) visuospatial explanation. We position a working model for deep semantics by highlighting select recent / closely related works from IJCAI, AAAI, ILP, and ACS. We posit that human-centred, explainable visual sensemaking necessitates both high-level semantics and low-level visual computing, with the highlighted works providing a model for systematic, modular integration of diverse multifaceted techniques developed in AI, ML, and Computer Vision.
|
|
| 3. |
- Eilertsen, Gabriel, 1984-, et al.
(författare)
-
Classifying the classifier : dissecting the weight space of neural networks
- 2020
-
Ingår i: Proceedings of the 24th European Conference on Artificial Intelligence (ECAI 2020). - : IOS PRESS. - 9781643681016 ; , s. 1119-1126
-
Konferensbidrag (refereegranskat)abstract
- This paper presents an empirical study on the weights of neural networks, where we interpret each model as a point in a high-dimensional space – the neural weight space. To explore the complex structure of this space, we sample from a diverse selection of training variations (dataset, optimization procedure, architecture,etc.) of neural network classifiers, and train a large number of models to represent the weight space. Then, we use a machine learning approach for analyzing and extracting information from this space. Most centrally, we train a number of novel deep meta-classifiers withthe objective of classifying different properties of the training setup by identifying their footprints in the weight space. Thus, the meta-classifiers probe for patterns induced by hyper-parameters, so that we can quantify how much, where, and when these are encoded through the optimization process. This provides a novel and complementary view for explainable AI, and we show how meta-classifiers can reveal a great deal of information about the training setup and optimization, by only considering a small subset of randomly selected consecutive weights. To promote further research on the weight space, we release the neural weight space (NWS) dataset – a collection of 320K weightsnapshots from 16K individually trained deep neural networks.
|
|
| 4. |
- Morveli-Espinoza, Mariela, et al.
(författare)
-
Towards an Imprecise Probability Approach forAbstract Argumentation
- 2020
-
Ingår i: ECAI 2020. - : IOS Press. - 9781643681009 - 9781643681016 ; , s. 2921-2922
-
Konferensbidrag (refereegranskat)abstract
- In some abstract argumentation framework (AAF), arguments have a degree of uncertainty, which impacts on the degreeof uncertainty of the extensions obtained under a semantics. In these approaches, both the uncertainty of the arguments and of the extensions are modeled by means of precise probability values. However, in many real life situations the exact probabilities values are unknownand sometimes there is a need for aggregating the probability valuesof different sources. In this paper, we tackle the problem of calculat-ing the degree of uncertainty of the extensions considering that theprobability values of the arguments are imprecise
|
|
| 5. |
- Singh, Avinash, 1986-, et al.
(författare)
-
A Fuzzy Inference System for a Visually Grounded Robot State of Mind
- 2020
-
Ingår i: ECAI 2020. - : IOS Press. - 9781643681009 - 9781643681016 ; , s. 2402-2409
-
Konferensbidrag (refereegranskat)abstract
- In order for robots to interact with humans on real-world scenarios or objects, these robots need to construct a representa- tion (‘state of mind’) of these scenarios that a) are grounded in the robots’ perception and b) ideally should match human understand- ing and concepts. Using table-top settings as scenario, we propose a framework that generates a robot’s ’‘state of mind’ by extracting the objects on the table along with their properties (color, shape and texture) and spatial relations to each other. The scene as perceived by the robot is represented in a dynamic graph in which object at- tributes are encoded as fuzzy linguistic variables that match human spatial concepts. In particular, this paper details the construction of such graph representations by combining low-level neural network- based feature recognition and a high-level fuzzy inference system. Using fuzzy representations allows for easily adapting the robot’s original scene representation to deviations in properties or relations that emerge in language descriptions given by humans viewing the same scene. The framework is implemented on a Pepper humanoid robot and has been evaluated using a data set collected in-house.
|
|
| 6. |
- Suchan, Jakob, et al.
(författare)
-
Driven by Commonsense : On the Role of Human-Centred Visual Explainability for Autonomous Vehicles
- 2020
-
Ingår i: ECAI 2020. - : IOS Press. - 9781643681009 - 9781643681016 ; , s. 2939-2940
-
Konferensbidrag (refereegranskat)abstract
- Within the autonomous driving domain, there is now a clear need and tremendous potential for hybrid solutions (e.g., integrating semantics, learning, visual computing) towards fulfilling essential legal and ethical responsibilities involving explainability (e.g., for diagnosis), human-centred AI (e.g., interaction design), and industrial standardisation (e.g, pertaining to representation, realisation of rules & norms). In these contexts, this highlight paper positions recent research from IJCAI 2019 [4] aimed at advancing human-centred AI principles in the backdrop of the autonomous driving application domain. From a technical viewpoint, the highlighted research provides a model for advancing the state of the art in reasoning about space and motion, combining reasoning and learning, non-monotonic reasoning, and computational modelling of high-level visuospatial commonsense. In addition to demonstrating the significance of integrated vision and semantics solutions in autonomous driving, we also highlight open questions emphasising the need for interdisciplinary mixed-methods research-involving AI, Psychology, HCI- to better appreciate the complexity and spectrum of varied human-centred challenges in diverse naturalistic driving situations.
|
|
| 7. |
- Tomic, Stevan, 1981-, et al.
(författare)
-
Learning Normative Behaviors through Abstraction
- 2020
-
Ingår i: ECAI 2020. - : IOS Press. - 9781643681009 - 9781643681016 ; , s. 1547-1554
-
Konferensbidrag (refereegranskat)abstract
- Future robots should follow human social norms to be useful and accepted in human society. In this paper, we show how prior knowledge about social norms, represented using an existing normative framework, can be used to (1) guide reinforcement learning agents towards normative policies, and (2) re-use (transfer) learned policies in novel domains. The proposed method is not dependent on a particular reinforcement learning algorithm and can be seen as a means to learn abstract procedural knowledge based on declarative domain-independent semantic specifications.
|
|
