Research in artificial intelligence and law goes back approximately 40 years. It remains largely based on formal logic, including non-monotonic logic, case-based reasoning, and logic programming. However, some researchers in and practitioners of law have argued in favor of quantitative approaches (e.g. probability) to account for uncertainties in legal arguments. Other researchers have pointed some of the shortcomings of the current artificial intelligence and law research, e.g. inability to take context into account. At the same time, machine learning has made huge inroads in many different fields and applications, and therefore, the question is whether machine learning has anything to offer to the theory, and, equally important, the practice of law. As a position paper, this is a preliminary study towards the exploration of a synergistic integration of current artificial intelligence approaches in law, with machine learning approaches. It puts forward the idea that formal, logic-based approaches, currently very popular the Artificial Intelligence & Law research, could benefit from an extension with a machine learning component, and discusses some ways in which machine learning could be integrated into these approaches.
In applying machine learning to malware identification, different types of features have proven to be successful. These features have also been tested with different kinds of classification methodologies and have had varying degrees of success. Every time a new machine learning methodology is introduced for classifying malware, there is the potential for increasing the overall quality of malware classification in the field. Even new classifiers with the same accuracy as those used previously can be combined using one of a few different ensemble techniques sharpen the classification and raise the accuracy to new heights. For our purposes, we have attempted to create a coalition of classifiers which each use different features. These classifiers when trained, provide multiple angles to the same problem and can be used to test ensemble techniques. Eventually, such an ensemble of individual malware classifiers could create a highly precise means of picking out malware from other software. Specifically, we have created a convolutional neural network which processes byte data as an image, and a deep feed forward neural network which utilizes opcode N-gram features. Both of these classifiers, while not perfect, provide a significant level of classification. They achieve this independently of one another, and when combined, they each contribute enough to improve the final accuracy. The majority of the effort in this research was placed on gathering the N-gram features, a time and resource intensive process. Tinkering with the parameters or structure of classifiers could provide further improvements to the system.
Deception is a core component of human interaction and reasoning, and despite its negative connotation, it can be used in positive ways. We present our formalization behind strategic deception, one such potentially positive form of deception. We use the Cognitive Event Calculus (CEC) to model strategic deception, building on prior formalizations. First, we provide a brief overview of deception’s definitions within existing literature. Following this discussion, CEC is described and we present CEC-style inference rules for strategic deception. These rules and a positive motivating deception example are used to show how we can solve the problem of strategic deception. This proof is demonstrated both through application of our rules and by adapting our rules for MATR (Machina Arachne Tree-based Reasoner) to show how proving can be performed by automatic reasoners. Finally, we discuss what future steps can be taken with strategic deception.
Interactive partially observable Markov decision processes (I-POMDPs) provide a principled framework for planning and acting in a partially observable, stochastic and multi-agent environment. It extends POMDPs to multi-agent settings by including models of other agents in the state space and forming a hierarchical belief structure. In order to predict other agents' actions using I-POMDPs, we propose an approach that effectively uses Bayesian inference and sequential Monte Carlo sampling to learn others' intentional models which ascribe to them beliefs, preferences and rationality in action selection. Empirical results show that our algorithm accurately learns models of the other agent and has superior performance than methods that use subintentional models. Our approach serves as a generalized Bayesian learning algorithm that learns other agents' beliefs, strategy levels, and transition, observation and reward functions. It also effectively mitigates the belief space complexity due to the nested belief hierarchy.
The primary focus of this research is to introduce a method of measuring an individual’s IQ by analyzing the vocabulary in said individual’s writing. In this paper, we show that the ratio of SAT words in a dataset of writing samples is roughly normally distributed, though with an obvious left skew. We go on to show a method that can be used to calculate an individual’s IQ with this ratio and provide samples with measured accuracy. The conclusion suggests ways to increase accuracy in order to further develop the research along with applications of doing so.
Full-duplex conversation where everybody can talk and hear at the same time is made possible by typed-chat computer-mediated communication. This experiment examines typing logs from students engaging in overlapping dialogue chat in small-group problem-solving sessions. When students are typing in the presence of overlapping dialogue there are measurable differences in their typing behavior. A difference measured here is text-deletion behavior. Deletions increase in the simultaneous typing regime. The reasons for this difference remain to be explored.
With the increasing volume of data, there is a growing need for dynamic data visualization to help reveal instant changes in data patterns. There exist many commercial visualization tools, but traditional scholars are often disengaged from the tool development process; thus, the choice of functionalities is contingent upon tool developers whose choice may not fit the end-users. This collaboration, however, has a potential in bridging the gap between traditional scholars, who are more interested in sense-making of the text than in the tools, and the data scientists, who are more interested in the tools than in the substance, but must still contextualize the outcomes. Until recently, this collaborative process was hindered by the complexity of customization procedures and technological hurdles imposed on users with new installations. With the advent of reactive web frameworks, such as Shiny, the user-driven customization becomes not only feasible, but also essential to advance scientific research. In this paper, we demonstrate a collaborative e↵ort between learned scholars and tool developers, allowing for a computational and humanistic fusion.
IoT technology has been recently adopted in the healthcare system to collect Electrocardiogram (ECG) signals for heart disease diagnosis and prediction. However, noises in collected ECG signals make the diagnosis and prediction system unreliable and imprecise. In this work, we have proposed a new lightweight approach to removing noises in collected ECG signals to perform precise diagnosis and prediction. First, we have used a revised Sequential Recursive (SR) algorithm to transform the signals into digital format. Then, the digital data is proceeded using a revised Discrete Wavelet Transform (DWT) algorithm to detect peaks in the data to remove noises. Finally, we extract some key features from the data to perform diagnosis and prediction based on a feature dataset. Redundant features are removed by using Fishers Linear Discriminant (FLD). We have used an ECG dataset from MIT-BIH (PhisioNet) to build a knowledge-base diagnosis features. We have implemented a proof-ofconcept system that collects and processes real ECG signals to perform heart disease diagnosis and prediction based on the built knowledgebase.
The raw Urdu corpus comprises of irregular and large sentences which need to be properly segmented in order to make them useful in Natural Language Engineering (NLE). This makes the Compound Sentences Segmentation (CSS) timely and vital research topic. The existing online text processing tools are developed mostly for computationally developed languages such as English, Japanese and Spanish etc., where sentence segmentation is mostly done on the basis of delimiters. Our proposed approach uses special characters as sentence delimiters and computationally extracted sentence-endletters and sentence-end-words as identifiers for segmentation of large and compound sentences. The raw and unannotated input text is passed through preprocessing and word segmentation. Urdu word segmentation itself is a complex task including knotty problems such as space insertion and space deletion etc. Main and subordinate clauses are identified and marked for subsequent processing. The resultant text is further processed in order to identify, extract and then segment large as well as compound sentences into regular Urdu sentences. Urdu computational research is in its infancy. Our work is pioneering in Urdu CSS and results achieved by our proposed approach are promising. For experimentation, we used a general genre raw Urdu corpus containing 2616 sentences and 291503 words. We achieved 34% improvement in reduction of average sentence length from 111 w/s to 38 w/s (words per sentence). This increased the number of sentences by almost three times to 7536 shorter and computationally easy to manage sentences. Resultant text reliability and coherence are verified by Urdu language experts.
This paper presents a computational framework for identity and is particularly focused on identifying the culprit in a crime-scene investigation. A case is conceptualized as a constellation of situations in the sense of Barwise’s situation theory. Data on a case is stored as RDF triples in a triple store. Several relevant OWL ontologies have been developed and supplemented with SWRL rules. Uncertainty and combining levels of (possibly conflicting) evidence are handled with Dempster-Shafer theory. A webpage is being developed to make available to students of criminal justice the results of our work. The user will be able to query about evidence and follow how it accrues to various hypotheses.
In order to protect valuable computer systems, network data needs to be analyzed and classified so that possible network intrusions can be detected. Machine learning techniques have been used to classify network data. For supervised machine learning methods, they can achieve high accuracy at classifying network data as normal or malicious, but they require the availability of fully labeled data. Semi-supervised machine learning methods, however, can use a small number of labeled examples and train a large number of examples without label. In this research, we explore the use of semi-supervised Random Forest in classifying network data and intrusion detection. It was used to classify the Third International Knowledge Discovery and Data Mining Tools Competition dataset (KDD 1999) and the result were compared with the results of using the supervised methods of Random Forest. The results were also compared with those using ladder network, an approach which combines unsupervised neural networks, in classifying KDD 1999.
