-
Understanding Boolean Function Learnability on Deep Neural Networks: PAC Learning Meets Neurosymbolic Models
Authors:
Marcio Nicolau,
Anderson R. Tavares,
Zhiwei Zhang,
Pedro Avelar,
João M. Flach,
Luis C. Lamb,
Moshe Y. Vardi
Abstract:
Computational learning theory states that many classes of boolean formulas are learnable in polynomial time. This paper addresses the understudied subject of how, in practice, such formulas can be learned by deep neural networks. Specifically, we analyze boolean formulas associated with model-sampling benchmarks, combinatorial optimization problems, and random 3-CNFs with varying degrees of constr…
▽ More
Computational learning theory states that many classes of boolean formulas are learnable in polynomial time. This paper addresses the understudied subject of how, in practice, such formulas can be learned by deep neural networks. Specifically, we analyze boolean formulas associated with model-sampling benchmarks, combinatorial optimization problems, and random 3-CNFs with varying degrees of constrainedness. Our experiments indicate that: (i) neural learning generalizes better than pure rule-based systems and pure symbolic approach; (ii) relatively small and shallow neural networks are very good approximators of formulas associated with combinatorial optimization problems; (iii) smaller formulas seem harder to learn, possibly due to the fewer positive (satisfying) examples available; and (iv) interestingly, underconstrained 3-CNF formulas are more challenging to learn than overconstrained ones. Such findings pave the way for a better understanding, construction, and use of interpretable neurosymbolic AI methods.
△ Less
Submitted 17 November, 2022; v1 submitted 12 September, 2020;
originally announced September 2020.
-
Maximally Divergent Intervals for Anomaly Detection
Authors:
Erik Rodner,
Björn Barz,
Yanira Guanche,
Milan Flach,
Miguel Mahecha,
Paul Bodesheim,
Markus Reichstein,
Joachim Denzler
Abstract:
We present new methods for batch anomaly detection in multivariate time series. Our methods are based on maximizing the Kullback-Leibler divergence between the data distribution within and outside an interval of the time series. An empirical analysis shows the benefits of our algorithms compared to methods that treat each time step independently from each other without optimizing with respect to a…
▽ More
We present new methods for batch anomaly detection in multivariate time series. Our methods are based on maximizing the Kullback-Leibler divergence between the data distribution within and outside an interval of the time series. An empirical analysis shows the benefits of our algorithms compared to methods that treat each time step independently from each other without optimizing with respect to all possible intervals.
△ Less
Submitted 21 October, 2016;
originally announced October 2016.
-
On Using Synthetic Social Media Stimuli in an Emergency Preparedness Functional Exercise
Authors:
Andrew Hampton,
Shreyansh Bhatt,
Alan Smith,
Jeremy Brunn,
Hemant Purohit,
Valerie L. Shalin,
John M. Flach,
Amit P. Sheth
Abstract:
This paper details the creation and use of a massive (over 32,000 messages) artificially constructed 'Twitter' microblog stream for a regional emergency preparedness functional exercise. By combining microblog conversion, manual production, and a control set, we created a web based information stream providing valid, misleading, and irrelevant information to public information officers (PIOs) repr…
▽ More
This paper details the creation and use of a massive (over 32,000 messages) artificially constructed 'Twitter' microblog stream for a regional emergency preparedness functional exercise. By combining microblog conversion, manual production, and a control set, we created a web based information stream providing valid, misleading, and irrelevant information to public information officers (PIOs) representing hospitals, fire departments, the local Red Cross, and city and county government officials. PIOs searched, monitored, and (through conventional channels) verified potentially acionable information that could then be redistributed through a personalized screen name. Our case study of a key PIO reveals several capabilities that social media can support, including event detection, the distribution of information between functions within the emergency response community, and the distribution of messages to the public. We suggest that training as well as information filtering tools are necessary to realize the potential of social media in both emergencies and exercises.
△ Less
Submitted 2 March, 2015;
originally announced March 2015.
