University College Cork


Partial (Neighbourhood) Singleton Arc Consistency for Constraint Satisfaction Problems

AAAI Conferences

Algorithms based on the general idea of singleton arc consistency (SAC) show considerable promise for improving backtrack search algorithms for constraint satisfaction problems (CSPs). The drawback is that even the most efficient efficient of them is still comparatively expensive. Even when limited to preprocessing, they give overall improvement only when problems are quite difficult to solve with more typical procedures such as maintained arc consistency (MAC). The present work examines a form of partial SAC and neighbourhood SAC (NSAC) in which a subset of the variables in a CSP are chosen to be made SAC-consistent or neighbourhood-SAC-consistent. It is shown that, using the proper procedures, partial (N)SAC is associated with a unique fixpoint for any given subset of variables. Various heuristic strategies for choosing the designated subset are described and tested, in particular a strategy of choosing by constraint weight after random probing. Experimental results justify the claim that these methods can be nearly as effective as full (N)SAC in terms of values discarded while greatly reducing the effort required.


Introducing Hypertension FACT: Vital Sign Ontology Annotations in the Florida Annotated Corpus for Translational Science

AAAI Conferences

We introduce the Florida Annotated Corpus for Translational Science (FACTS), which currently consists of 20 case reports about hypertension that we annotated with Vital Sign Ontology (VSO) classes. We describe the annotation method, the annotation results, interannotator agreement measure, and the availability of the corpus and supporting tools for annotating corpora with OWL ontologies. We also discuss issues and limitations of VSO for annotating vital sign data in case reports.


Hicks

AAAI Conferences

We introduce the Florida Annotated Corpus for Translational Science (FACTS), which currently consists of 20 case reports about hypertension that we annotated with Vital Sign Ontology (VSO) classes. We describe the annotation method, the annotation results, interannotator agreement measure, and the availability of the corpus and supporting tools for annotating corpora with OWL ontologies. We also discuss issues and limitations of VSO for annotating vital sign data in case reports.


Wallace

AAAI Conferences

Algorithms based on the general idea of singleton arc consistency (SAC) show considerable promise for improving backtrack search algorithms for constraint satisfaction problems (CSPs). The drawback is that even the most efficient efficient of them is still comparatively expensive. Even when limited to preprocessing, they give overall improvement only when problems are quite difficult to solve with more typical procedures such as maintained arc consistency (MAC). The present work examines a form of partial SAC and neighbourhood SAC (NSAC) in which a subset of the variables in a CSP are chosen to be made SAC-consistent or neighbourhood-SAC-consistent. It is shown that, using the proper procedures, partial (N)SAC is associated with a unique fixpoint for any given subset of variables. Various heuristic strategies for choosing the designated subset are described and tested, in particular a strategy of choosing by constraint weight after random probing. Experimental results justify the claim that these methods can be nearly as effective as full (N)SAC in terms of values discarded while greatly reducing the effort required.


Denoising Dictionary Learning Against Adversarial Perturbations

AAAI Conferences

We propose denoising dictionary learning (DDL), a simple yet effective technique as a protection measure against adversarial perturbations. We examined demising dictionary learning on MNIST andCIFAR10 perturbed under two different perturbation techniques, fast gradient sign (FGSM) and jacobian saliency maps (JSMA). We evaluated it against five different deep neural networks (DNN) representing the building blocks of most recent architectures indicating a successive progression of model complexity of each other. We show that each model tends to capture different representations based on their architecture. For each model we recorded its accuracy both on the perturbed test data previously misclassified with high confidence and on the denoised one after the reconstruction using dictionary learning. The reconstruction quality of each data point is assessed by means of PSNR (Peak Signal to Noise Ratio) and Structure Similarity Index (SSI). We show that after applying (DDL) the reconstruction of the original data point from a noisy sample results in a correct prediction with high confidence.


Mitro

AAAI Conferences

We propose denoising dictionary learning (DDL), a simple yet effective technique as a protection measure against adversarial perturbations. We examined demising dictionary learning on MNIST andCIFAR10 perturbed under two different perturbation techniques, fast gradient sign (FGSM) and jacobian saliency maps (JSMA). We evaluated it against five different deep neural networks (DNN) representing the building blocks of most recent architectures indicating a successive progression of model complexity of each other. We show that each model tends to capture different representations based on their architecture. For each model we recorded its accuracy both on the perturbed test data previously misclassified with high confidence and on the denoised one after the reconstruction using dictionary learning. The reconstruction quality of each data point is assessed by means of PSNR (Peak Signal to Noise Ratio) and Structure Similarity Index (SSI). We show that after applying (DDL) the reconstruction of the original data point from a noisy sample results in a correct prediction with high confidence.


Recommending from Experience

AAAI Conferences

In this paper we present RC, a context-driven recommender system that mines contextual information from user-generated reviews and makes recommendations based on the users' experiences. RC mines the contextual information from the user-generated reviews using a form of topic modeling. This means that, unlike other context-aware recommender systems, RC does not have a predefined set of contextual variables. After mining the contextual information, RC makes top-n recommendations using a Factorization Machine with the contextual topics as side information. Our experiments on two datasets of ratings and reviews show that RC has higher recall than a conventional recommender.


A Normative-Prescriptive-Descriptive Approach to Analyzing CSP Heuristics

AAAI Conferences

This paper presents a general framework for analyzing heuristics for constraint solving, including backtracking and arc consistency algorithms. It will emphasize heuristics for variable selection during search, since this is where major differences are found. In earlier work two basic approaches to this problem were developed.The first was a general theoretical framework for different types of heuristics, which characterized ideal performance so that the actual performance of heuristics could be compared to this standard. The second involved the discovery that, while there are a large number of features that can be used for heuristic decisions in variable ordering, differences in effectiveness boil down to only two basic "heuristic actions". The present paper applies basic ideas from decision analysis to characterize these two approaches to better understand their status and interrelations. It shows that the first is essentially a normative decision analysis, and that models of this sort imply general prescriptive principles (notably the Fail-First Principle). The second is concerned with descriptive models of actual performance.


Wallace

AAAI Conferences

This paper presents a general framework for analyzing heuristics for constraint solving, including backtracking and arc consistency algorithms. It will emphasize heuristics for variable selection during search, since this is where major differences are found. In earlier work two basic approaches to this problem were developed.The first was a general theoretical framework for different types of heuristics, which characterized ideal performance so that the actual performance of heuristics could be compared to this standard. The second involved the discovery that, while there are a large number of features that can be used for heuristic decisions in variable ordering, differences in effectiveness boil down to only two basic "heuristic actions". The present paper applies basic ideas from decision analysis to characterize these two approaches to better understand their status and interrelations. It shows that the first is essentially a normative decision analysis, and that models of this sort imply general prescriptive principles (notably the Fail-First Principle). The second is concerned with descriptive models of actual performance.


Explaining Ourselves: Human-Aware Constraint Reasoning

AAAI Conferences

Human-aware AI is increasingly important as AI becomes more powerful and ubiquitous. A good foundation for human-awareness should enable ourselves and our "AIs" to "explain ourselves" naturally to each other. Constraint reasoning offers particular opportunities and challenges in this regard. This paper takes note of the history of work in this area and encourages increased attention, laying out a rough research agenda.