Cognitive-Computing

Cognitive computing is the use of computerized models to synthetically simulate human thought processes in complex situations where the answers may be ambiguous and uncertain. The field overlaps with computational intelligence and involves many of the same underlying technologies to power cognitive applications, including fuzzy systems, neural networks and evolutionary computing as well as augmented, mixed and virtual reality. Our focus here is on explainable and trustworthy artificial intelligence as well as interactive and collaborative machine learning.

  • An explainable framework for predicting lung cancer genetic mutations using fuzzy few shot guessing

    his research proposes a framework for predicting and explaining lung cancer prognosis and genetic mutations using a few high-quality histologic images and a fuzzy few shot learning algorithm based on fuzzy clustering methods. The framework includes a Web Ontology Language (OWL 2.0) to create a tree of fuzzy and crisp relationships between the classes of lung cancers, histologic pattern growth, and genetic mutations. The results are explained to specialized physicians. The framework is benchmarked with state-of-the-art solutions like standard few shot learning and deep learning, comparing them in terms of images required for training, accuracy and speed of prediction, extrapolation to genetic mutations, and explainability of the results.

    Human-IST collaborators: Servio Lima, Luis Terán

    Partners/ External collaborators (companies): SOLCA Ecuador, NVIDIA

    Publications:

    ABSTRACT published 2019: Explainable neuro-fuzzy recurrent neural network to predict colo-rectal cancer with different timeframe data (EUSFLAT Chequia 2019) PAPER published 2020: A Proposal for an Explainable Fuzzy-based Deep Learning System for Skin Cancer Prediction (ICEDEG Argentina APRIL 2020)

    WORKING PAPERS:

    • Fuzzy membership as an alternative method to similarity functions applied to few shot learning
    • An explainable fuzzy ontology web language applied to lung cancer histologic patterns
  • Neuro-fuzzy explainable systems for medicine

    One of the reasons why deep neural networks haven´t been adopted pervasively in medicine is the lack of explainability that helps physicians to know why an algorithm yielded a prediction. It is well known that deep learning relies on data to learn, so if data is biased or with errors, prediction will be also with errors. Moreover, modern attacks aimed towards deep neural networks make these systems more unreliable. This is why the field of medicine needs explainable systems that help physicians determine that the reasons of the predictions were correct. Neuro-fuzzy systems are based on fuzzy logic for interpreting the predictions of neural networks using fuzzy rules. These rules can be converted to natural language by means of several techniques. Both, the fuzzy rules and the natural language can be analyzed by a human who can use his expert criteria to decide if those rules or natural language explanations are correct. A good consequence of this, is that the whole system can be maintained and corrected by expert criteria, while preserving the accuracy of deep neural systems.

    Human-IST collaborators: Servio Lima, Luis Terán, Edy Portmann

    Partners/ External collaborators (companies): Doctor Online

  • Fuzzy explainable self-deep learning recommender algorithms for social networks

    In mathematical modelling we come across two inconveniences. The first is caused by the excessive complexity of the situation being modeled, which leads to very complex models to be used in practice or NP-Complete problems, where there is no known way to find a solution quickly. The second inconvenience consists of the indeterminacy, which is represented by vagueness of its semantics and our capability to work with vague notions. Fuzzy mathematics applied to machine learning in the context of explainable systems for recommender systems, can bring a better approach to model mathematically complex situations in a more human understandable way, reducing the complexity and leading to a better understanding of the reasoning behind the algorithm decisions.

    This project aims to model mathematically, using our unique Fuzzy Recommender - Machine Learning approach, the human interaction in different social networks under the Beekeeper platform across several industries. Where those models are tolerant of imprecision, uncertainty, partial truth and approximation.

     

    Human-IST collaborators: Jose Mancera, Luis Terán, Edy Portmann
    Partners/ External collaborators (companies): Beekeeper

  • Smartifying the Last-mile Delivery through Computational Intelligence

    This research work proposes an improvement of the first-try delivery by studying the behavior of traffic on the streets and customers' presence at home. In contrast to existing solutions, it is proposed working only with data that does not compromise the customers' privacy (i.e., avoiding the use of tracking data) and getting insights about traffic characteristics without the need of deploying a large number of vehicles or expensive sensors. The main goal is to provide a delivery route plan to delivery teams and route planners, which allows finishing the distribution of the parcels in the least amount of time, while being able to effectively deliver the highest amount of them. This will be translated into less resource consumption and a possible increased customer satisfaction.

    The aforementioned can be achieved through the use of computational intelligence (CI), which entails different methods, theories, and concepts aiming at bringing the abilities of computer systems closer to human cognition. CI provides ways of dealing with uncertainty and inaccurate data, that combined with a human-centric approach can allow developing a solution that truly adjusts to the needs of the users, being customers and the delivery companies’ in this case.

    Human-IST collaborators: Jhonny Pincay, Edy Portmann, Luis Terán

    Partners/ External collaborators (companies): Swiss Post, Via Suisse, Secretariat of High Education, Science, Technology and Innovation (SENESCYT)

  • Phenotropic Interaction

    To ensure better interactions between systems of various nature, a bio-inspired way of handling exchanges is being studied. The basic idea of Phenotropic Interaction is that of handling interactions with flexibility, instead of following strict and predefined protocols, by having all involved systems adapt to the others, in such a way that interactions become more and more flawless and precise over time. This kind of flexibility brings robustness and customizability of actions and interactions. Phenotropics can be applied for example in Human-Computer Interaction, but also in communication between pieces of software or computer hardware.

    Human-IST collaborators: Moreno Colombo, Edy Portmann, Denis Lalanne, Sara D’Onofrio

    Partners/ External collaborators (companies): Swiss Post, Swiss National Science Foundation (SNF)

  • Fuzzy Management System in Customer Services

    To help agents inside the Contact Center Post improving his performance and knowledge, an information and knowledge management (IKM) system will be implemented to provide: (1) right information to collaborators in order to answer clients’ questions in a short time, (2) a list of experts who can help him out if he can not handle a special case and (3) recommendations related to his search terms and new arrivals information. Fuzzy databases and techniques will be used in this IKM system to give more precise recommendations.

    Human-IST collaborators: Minh Tue Nguyen, Edy Portmann

    Partners/ External collaborators (companies): Swiss Post

  • Fuzzy Leadership in People Development

    To help the Swiss Post’s collaborators take ownership of their professional development a guiding virtual assistant will encompass 30 years of psychology and neuroscience advances and foster the individual’s self-authorship through chatbot-like interactions.

    Human-IST collaborators: Timo Schuler, Edy Portmann

    Partners/ External collaborators (companies): Swiss Post

  • Internet of Postal Things (IoPT)

    In this project, we are experimenting with innovative postal services in the field of Internet of Things (IoT) sensors and incidental business models. The basic notion is to make better use of the physical Web Swiss Posts owns and to provide analyzed data of the IoT-sensors to customers or citizens. The project includes technical challenges (i.e. LoRa WAN, new kind of sensors, etc.) as well as business challenges (i.e. new business models as Freemium services, etc.), which should be explored experimentally, as they go hand-in-hand and therefore need to be addressed holistically (i.e. applying action-design-research). It is a design-science research project, where sociotechnical issues as pioneering citizen-centered Smart City services are addressed.

    Human-IST Collaborators: Edy Portmann

    External Collaborators: Swiss Post

    Start date: September 2017

  • An AI Hierarchy of Needs for Swiss Post Services

    This project looks at postal data (e.g. scans, logins, sensor data, user generated content, etc.) to feed smart systems, which could provide users (e.g. citizens) with personalized and contextual information (e.g. within their city). The data come from different parts within Swiss Post (e.g. PostBus, PostFinance, PostGroup, etc.) as well as in different forms and granularity (e.g. ambiguous sensor data, natural language sources, etc.). The quest of this design-science research is, to design and engineer AI-systems, which offer value to the users (e.g. through providing users/citizens personalized information in the right context). The basic challenges thereby range from collecting data via exploring them through to self-learning systems, which may communicate with users/citizens (e.g. in natural language and with visualization).

    Human-IST Collaborators: Edy Portmann

    External Collaborators: Swiss Post

    Start date: September 2017

  • Computing with Words and Perceptions

    Computing with Words and Perceptions is a relatively new method of computing for computer systems. As words occur in the “natural language” and not in the numerical format, they are imprecise and cannot be processed by traditional systems. Because of this, a lot of relevant information gets lost. In order to change that, the team of the Human-ist examines linguistic computing and tries to create intelligent and smart computer systems. With the help of Computing with Words and Perceptions, new ways of computing should be developed which make the transformation of imprecise information into computer-readable data possible so that they can subsequently be applied in computations. It should be possible to include all necessary information in decision-making processes. It is thus the aim to make today’s processing of information more efficient such that the interaction between humans and computer systems can be optimized.

    Human-IST Collaborators: Edy Portmann, Sara D’Onofrio

    External Collaborators: Swiss Post

    Start date: September 2017