ICAISC 2019

Great progress has been made in recent years in methods of measurement and analysis of neuroimaging and electrophysiological data. Connectomes define anatomical organization of brain networks, and functional dynamical states selectively activate such networks to process information. Network science helps to analyze the whole-brain functional network dynamics, in some cases showing how to transform it into mental activity. Neural dynamics depends on the cognitive load and the history of brain activity, reorganizing functional networks, making the mental states context and experience-dependent.
Extracting "fingerprints" of active brain regions or subnetworks from fMRI and EEG data opens the path to many applications: reliable brain-computer interfaces, diagnostic methods in neuropsychiatry, therapeutic interventions using neuromodulation, optimization of brain processes through neurofeedback or behavioral procedures, and interpretation of some mental states. I will present several approaches to identify specific brain activity from high density EEG signals, based on spectral fingerprints, source reconstruction and localization, and functional network science analysis. There is hope that such methods may have wide applications in psychiatry, providing objective methods to characterise mental states and diagnose diseases.

Literature:
1. Duch W, Multi-level Explanations in Neuroscience I: From genes to subjective experiences. Acta Physica Polonica B 49(12), 1981-2010, 2018
2. Finc K, Bonna K, Lewandowska M, Wolak T, Nikadon J, Dreszer J, Duch W, Kühn S. Whole-brain functional network modularity and efficiency changes related to cognitive effort. Human Brain Mapping 38(7), 3659–3674, 2017
3. Gravier A, Quek H.C, Duch W, Abdul Wahab, Gravier-Rymaszewska J. Neural network modelling of the influence of channelopathies on reflex visual attention. Cognitive Neurodynamics 10(1), 49-72, 2016

In this talk, I would cover the value of tracking social network interactions, in particular tweeter, for various health-related topics. This type of analysis would allow us to recognize the perception of tweeter users in various locations while tracking changes over time. This work is part of a larger project to have multi-lingual analysis and to integrate health perception with health-related data and literature. The techniques used include NLP and ANN. A system that has been developed used a combination of unsupervised and supervised algorithms to track the trends of health-related topics in social media. The system captures the correlation of words based on the context to detect the trending topics. The supervised algorithm implemented is a Convolutional Neural Network (CNN) in conjunction with the Word2Vect model to classify and label new tweets, assigning a feedback to the topic models. Our results are compared with two state of the art techniques demonstrating an advantage that can be leveraged for further improvements.

We argue that the model-free Exploratory Data Analysis or Data Mining can be used to advantage as the first stage of a two-stage data-driven causal reasoning process. We illustrate this by example - why our Monte Carlo Feature Selection and Interdependencies Discovery algorithm is a useful tool for such two-stage approach.

Data clustering is a popular unsupervised pattern classification technique that is used for partitioning a given data set into homogeneous groups based on some similarity/dissimilarity metric. There are several single objective clustering algorithms available and popular in the literature including KMean and FCM. Although both KMean and FCM are very popular algorithms, they may stuck to local optima depending on the choice of initial cluster centers. In this lecture first we will demonstrate how Metaheuristic techniques can be used to solve the problem of KMean/FCM. Result will be demonstrated for pixel classification of satellite images.

In the second part of the lecture we will discuss more on Multiobjective clustering, in which multiple objective functions are simultaneously optimized. Selecting one solution from the set of Pareto Optimal solutions is always a critical issue. We will also discuss how machine learning techniques like Support Vector Machine (SVM) can be used to combine the Pareto Optimal solutions to evolve a better solution. The result will be demonstrated for classification of Micro Array Gene Expression Data.

The talk would cover a few AI-based business applications for transforming data into decisions, based on work done for three companies (NuTech Solutions, SolveIT Software, and Complexica) over the last 20 years. A few general concepts (Adaptive Business Intelligence, Travelling Thief Problem, Larry – the Digital Analyst) would be discussed and illustrated by a few examples. The final part of the talk would present Complexica’s approach for increasing revenue, margin, and customer engagement through automated analysis.

In retrospect, over years Artificial Intelligence (AI) has exhibited facets of qualitative (symbolic) and sub-symbolic (numeric) knowledge representation, reasoning, and modeling. At the symbolic end of the spectrum of AI constructs, it becomes imperative to form symbols so that they are well aligned (grounded) with real-world experimental data and in this way become a sound abstraction and perception of the world.

In this study, we advocate that information granules, both in terms of their conceptual underpinnings and associated design constructs, offer a required environment supporting a transition from data to a collection of symbols and delivering a required level of abstraction.

The overall functional schemed outline here can be concisely portrayed in the form
data -› structure (clusters) -› granular clusters -› granular schemes of modeling and reasoning
where a buildup of information granules supports generative and discriminative aspects of clusters regarded as generic constructs. While clustering algorithms give rise to numeric constructs (say, prototypes and partition matrices, etc.) and with this regard exhibit a long tradition, a paradigm shift becomes inevitable. We propose here an attractive alternative of symbolic (qualitative) characterization of information granules (clusters), which supports higher levels of interpretability and offers insights into aspects of conceptual stability of structural findings. It also links a level of information granularity (abstraction) with the stability of symbols and emphasizes the duality of interpretation of information granules as exhibiting their duality of symbolic (syntactic) and semantic nature. Those aspects are directly aligned with qualitative modeling and qualitative reasoning researched intensively in the AI domain. We discuss a direction of research aimed at building optimal information granules and their abilities to represent the data. It is argued that when coping with distributed environment (say, multiagent systems), information granules of higher type and higher order become of interest.

Shannon's information theory has served as a bedrock for advances in communication and storage systems over the past five decades. However, this theory does not handle well higher order structures (e.g., graphs, geometric structures), temporal aspects (e.g., real-time considerations), or semantics. We argue that these are essential aspects of data and information that underly a broad class of current and emerging data science applications. In this talk, we present some recent results on structural and temporal information. We first show how to extract temporal information in dynamic networks (arrival of nodes) from its structure (unlabeled graphs). We then proceed to establish fundamental limits on information content for some data structures, and present asymptotically optimal lossless compression algorithms achieving these limits for various graph models.

We use these examples to motivate our view on data science. Our approach brings information to the center of the theoretical foundations of information and data science. In sum, we argue that the unifying theme of data science should be the triad: from data to information to knowledge.

Since the concept of Turing machine has been first proposed in 1936, the capability of machines to perform intelligent tasks went on growing exponentially. Artificial Intelligence (AI), as an essential accelerator, pursues the target of making machines as intelligent as human beings. It has already reformed how we live, work, learning, discover and communicate. In this talk, I will review our recent progress on AI by introducing some representative advancements from algorithms to applications, and illustrate the stairs for its realization from perceiving to learning, reasoning and behaving. To push AI from the narrow to the general, many challenges lie ahead. I will bring some examples out into the open, and shed lights on our future target. Today, we teach machines how to be intelligent as ourselves. Tomorrow, they will be our partners to step into our daily life.