Natural Language Processing (NLP) tasks, such as Named Entity Recognition (NER), involve an iterative process of model optimization to identify different types of words or semantic entities. This optimization to achieve a more precise model becomes computationally difficult as the number of iterations increase. The small datasets available for training typically limit the models. Adding iterations on such sets to further optimize the model can often cause over-fitting, which generally leads to reduced performance. Therefore, the choice of convergence criteria is a critical step in robust and accurate model building. We evaluate different convergence criteria in terms of their robustness, stopping threshold selection, and independence from the training data size and entity. The underlying framework employs a limited-memory Broyden-Fletcher-Goldfarb-Shanno (L-BFGS) parameter optimization in the context of Conditional Random Fields (CRF). This paper presents a convergence criterion for robust training irrespective of semantic types and data sizes with two-orders of magnitude reduction in stopping threshold for improved model accuracy and faster convergence. Additionally, we examine convergence with active learning to further reduce the training data and training time.

Conference and special track committees responsible for the 2009 FLAIRS conference.

List of officers of the FLAIRS society.

Many reinforcement learning exploration techniques are overly optimistic and try to explore every state. Such exploration is impossible in environments with the unlimited number of states. I propose to use simulated exploration with an optimistic model to discover promising paths for real exploration. This reduces the needs for the real exploration.

The emerging Web of Data utilizes the web infrastructure to represent and interrelate data. The foundational standards of the Web of Data include the Uniform Resource Identifier (URI) and the Resource Description Framework (RDF). URIs are used to identify resources and RDF is used to relate resources. While RDF has been posited as a logic language designed specifically for knowledge representation and reasoning, it is more generally useful if it can conveniently support other models of computing. In order to realize the Web of Data as a general-purpose medium for storing and processing the world's data, it is necessary to separate RDF from its logic language legacy and frame it simply as a data model. Moreover, there is significant advantage in seeing the Semantic Web as a particular interpretation of the Web of Data that is focused specifically on knowledge representation and reasoning. By doing so, other interpretations of the Web of Data are exposed that realize RDF in different capacities and in support of different computing models.

Graded modal logic is the formal language obtained from ordinary (propositional) modal logic by endowing its modal operators with cardinality constraints. Under the familiar possible-worlds semantics, these augmented modal operators receive interpretations such as "It is true at no fewer than 15 accessible worlds that...", or "It is true at no more than 2 accessible worlds that...". We investigate the complexity of satisfiability for this language over some familiar classes of frames. This problem is more challenging than its ordinary modal logic counterpart--especially in the case of transitive frames, where graded modal logic lacks the tree-model property. We obtain tight complexity bounds for the problem of determining the satisfiability of a given graded modal logic formula over the classes of frames characterized by any combination of reflexivity, seriality, symmetry, transitivity and the Euclidean property.

This paper further introduces and formalizes a novel concept of self-forensics for automotive vehicles, specified in the Forensic Lucid language. We argue that self-forensics, with the forensics taken out of the cybercrime domain, is applicable to "self-dissection" of intelligent vehicles and hardware systems for automated incident and anomaly analysis and event reconstruction by the software with or without the aid of the engineering teams in a variety of forensic scenarios. We propose a formal design, requirements, and specification of the self-forensic enabled units (similar to blackboxes) in vehicles that will help investigation of incidents and also automated reasoning and verification of theories along with the events reconstruction in a formal model. We argue such an analysis is beneficial to improve the safety of the passengers and their vehicles, like the airline industry does for planes.

We present a straightforward embedding of quantified multimodal logic in simple type theory and prove its soundness and completeness. Modal operators are replaced by quantification over a type of possible worlds. We present simple experiments, using existing higher-order theorem provers, to demonstrate that the embedding allows automated proofs of statements in these logics, as well as meta properties of them.

This paper presents the current state of a work in progress, whose objective is to better understand the effects of factors that significantly influence the performance of Latent Semantic Analysis (LSA). A difficult task, which consists in answering (French) biology Multiple Choice Questions, is used to test the semantic properties of the truncated singular space and to study the relative influence of main parameters. A dedicated software has been designed to fine tune the LSA semantic space for the Multiple Choice Questions task. With optimal parameters, the performances of our simple model are quite surprisingly equal or superior to those of 7th and 8th grades students. This indicates that semantic spaces were quite good despite their low dimensions and the small sizes of training data sets. Besides, we present an original entropy global weighting of answers' terms of each question of the Multiple Choice Questions which was necessary to achieve the model's success.

Multi-instance learning attempts to learn from a training set consisting of labeled bags each containing many unlabeled instances. Previous studies typically treat the instances in the bags as independently and identically distributed. However, the instances in a bag are rarely independent, and therefore a better performance can be expected if the instances are treated in an non-i.i.d. way that exploits the relations among instances. In this paper, we propose a simple yet effective multi-instance learning method, which regards each bag as a graph and uses a specific kernel to distinguish the graphs by considering the features of the nodes as well as the features of the edges that convey some relations among instances. The effectiveness of the proposed method is validated by experiments.