Table of Contents
- ADVANTAGES
- DEVELOPMENT LIFE CYCLE
- LEGAL ASPECTS AND PRIVACY
- CHALLENGES
- AFFORDABLE ARTIFICIAL INTELLIGENCE
- DATA MODELING AND INTEROPERABILITY
- READY TO USE DATA ANALYTICS
- Paolo Nesi
Descriptive, prescriptive and predictive solutions have been supplied for many years, from statistic, operating research, and regressive models. In most cases, they have not been capable to provide satisfactory results for their limitation of discovering/modelling complex functions and relationships. Early neural networks and machine learning solutions were mainly black box generating scepticism over their applicability in critical situations in which one should be requested to just trust them as black box. On the other hand, ethical aspects (on data and processes) are very sensitive and a wrong assumption in taking data and/or setting up solutions may lead to biased results/suggestions, which may correspond discriminations and may lead to unforeseen costs. This was a real concern for former solutions and for the first generations of machine learning.
The IoT combined with Big Data are enabling a large number of data analytics. Big Data Analytic with Machine Learning, ML, and Artificial Intelligence, AI, may play a strong role in leveraging businesses and solutions providing reliable predictions, prescriptions, early warning, classifications, detections, suggestions, etc. In substance, these technologies can lead to reduce costs and increase the efficiency of processes, being them business or production processes. This also implies to add value to big data collected, providing from them hints, strategies, mitigations, and discovering information and implications never detected before, that implies to learn some aspects of the phenomena up to now only observed and measured.
The applications can be in almost any domain of smart city and industry: mobility, health, energy, environment, waste, chemistry, manufactory, delivering, agriculture, boating, etc.
ADVANTAGES
The resulting advantages can be for final users, for decision-makers and for the city/companies. Examples can be found in the:
- parking prediction to reduce the social cost of looking for parking: reduction of fuel consumption, reduction of produced as NO2 for the whole community, etc.;
- optimization of waste collection that reduces costs by minimising the number of required trucks/trips and prevents waste from escaping bins through the use of predictive analytics. It is an advantage for the quality of life of city users, and reduces administrative costs;
- optimized positioning of bus stops, increasing the quality of transport experience and avoiding overcrowding, maximising efficiency, minimising costs, and decreasing the number of trips;
- predicting maintenance to reduce intervention costs, operating costs associated with unexpected faults and services/productions interruptions, avoiding damages provoked by faults, etc. This also implies a reduction of the costs of production improving efficiency and resilience;
- sharing service prediction to reduce the time need to get/find a suitable sharing vehicle. Thus, increasing the quality of service also increases its effectiveness;
- optimized routing based on present and predicted conditions, regarding traffic and maintenance works, for private drivers and city operators. Thus, for emergency services, reducing travel time, pollution, and the time required to reach the objective and perform services;
- assessing and predicting reputations of services, attractions, based on social media, with the purpose of enhancing the quality of services, reducing prices increase, and promoting alternative offers and solutions;
- understanding actual city usage conditions, and city users’ satisfactory and demanded services, etc. Increasing quality of services by solving the identified problems via collective intelligence.
- preparing the city and/or the industry to be more resilient to unexpected unknown events, natural or man-made disasters, by integrating simulations and ML/AI solutions enabling the what-if analysis in near real-time, increase resilience and capacity. Rapidly reacting to unanticipated, unforeseen events, so reducing the cost of recovery, which is the risk, and receiving advices to mitigate the risks and damages;
- Understanding the usage of city services to optimize energy and other resources needed for the expected serviceability and quality.
Present ML/AI/XAI solutions are capable to provide high precision in producing predictions, classification and prescriptions and thus they can actually support decision makers as one of the experts without the possible bias of humans. Thus, they should be treated as one of trusted expert and put the decision makers in the loop to collaborate with them to create tailored solutions, and strategies to mitigate and solve short- and log-terms problems, as well as to support processes of What-If analysis, which were previously developed solely through simulations. Decision-makers can get suggestions shortening boring activities and the AI can learn from the decision-makers’ objections improving progressively AI capabilities, taking into account additional aspects and elements. Extending this capability to allow entering in the loop also a community of users, it would result in the production of new data for training (suggestions produced by the users to the AI, as a collective intelligence) which can be actually improve the AI precision and capability in modelling the phenomena and providing effective suggestions. Some of these approaches fit in the reinforced learning and in the continuous learning techniques. Moreover, a relevant push has been provided by semantic reasoner tools, which started with the definition of ontologies and collecting data become actual expert systems which can be queries with semantic query language to perform inference and gest smart suggestions and results. Examples are the graph neural network, graph data stores and the spatial, temporal and entities reasoners.
In recent years, Big Data Analytics has rapidly evolved and started to be used in complex systems and not only to make direct predictions and prescriptions. Thus, decision-makers started to test the new solutions expecting to see them respecting the ethic (on data and processes), with aim of trusting them as their best experts. To this end, ML/AI trustworthy, Data Ethics and AI Ethics approaches have been created to be applied in the context of decision makers. Data Ethics refers to the aspects that may provoke a bias and ethical problems since the training for example, training the AI with biased data, unbalanced distribution of cases, etc. Moreover, specific ML/AI methodologies and solutions for Explainable Artificial Intelligence, XAI, are presently providing support in this direction since they are capable to explain the rationales behind the typical results provided (global explainable AI) and may provide specific description/rational for each result/suggestion provided (local explainable AI). XAI typically adds value to the provided decision producing hints and discovering implications and correlations never detected before. They are a source of information to train the decision-makers about what has been discovered to be relevant for the AI. On the other hand, the decision makers can also provide to the AI continuous and lifelong training inputs for improving the capabilities of the AI. Thus, in Europe (April 2019) and in recently in other countries normative are growing regarding the AI Ethics, proving guidelines on training AI for Ethics, thus for AI trustworthy, Data Ethics and AI Ethics.
DEVELOPMENT LIFE CYCLE
Nowadays is not enough to provide high precision in predictions, classification, counting, detecting, suggesting/prescripting and supporting decisions, the smart city and Industry 4.0 platforms must be capable to be intimately supported by AI/XAI methods to provide ethical, trustworthy and reliable solutions. And this integrated approach of ML/AI trustworthy, Data Ethics and AI Ethics must be enforced into the development process (e.g., data analysis, data selection, data ingestion, data ethic, data review, ML/AI selection, XAI selection).
LEGAL ASPECTS AND PRIVACY
Moreover, any ML/AI/XAI solution (including data ingestion, transformation, training, visualization, etc.) has to respect the data privacy, and thus has to be compliant with GDPR (General Data Protection Regulation of the European Commission) and/or similarly regulations in other non-European Countries (e.g., the California Consumer Privacy, CCPA). These aspects have to be addressed since the beginning of the above presented life cycle, when the data discovery and data ingestion are performed, and in particular in the data analysis phase. This also means that the solution has to respect the Data sovereignty for which the data are subject to the laws and governance structures of the nation where they were collected. Specific licenses can be modelled and the development tools enabling the development of AI must guarantee the Data sovereignty and GDPR. On the other hand, the long experience in data analytics also demonstrated that in a large number of cases, surrogated data can be found to substitute those that are protected or too private to be used. Specific techniques for anonymization preserving static validity may help in this sense and are getting a larger diffusion.
Thus, the first step to enforce the ML/AI in your process must pass from trusting and deep assessment / validation of the process and solutions proposed to verify its capabilities in being trustworthy, Ethics, and compliant with GDPR, AI Regulation of EU act, etc. in this phase, the methodologies of incidental finding have to be applied since, unexpected results and implications could be discovered, and in some cases could make evident the presence of hidden biased aspect on data and in the process.
CHALLENGES
Among the most relevant challenges for the city in the coming years we see the energy and ecological transitions KPI (key performance indicators), the SDG (Sustainable Development Goals) (see https://www.snap4city.org/776), and the push to more liveable cities according to 15 Min City Indexes and Driving urban transitions to a sustainable future, DUT. Sustainability and SDG are not something that one can solve with a single action. All the governs and institutions have identified hundreds of indicators which may influence the well-known SDG, DUT and 15MinCityIndex (https://www.snap4city.org/652).
DISIT Lab has developed the 15MinCityIndex, adopted in the whole Italian territory with the ENEL-X platform. And a number of solutions which can positively impact on the SDG. The 15MinCityIndex itself may be used to learn how to improve the SDG for each single micro area of the territory. https://www.snap4city.org/download/video/course/da/
AFFORDABLE ARTIFICIAL INTELLIGENCE
Moreover, the most recent techniques of transfer learning, augmented learning, etc., can drastically reduce the costs for setting up a ML/AI solution by starting from similar previous AI/XAI models also developed in similar contexts. The approach is reducing the amount of data and cases needed for training the ML/AL in the new conditions by exploiting pretrained models. These approaches imply that obtaining an affordable AI is not a matter of API or data models, while is more an issue of AI/ML/XAI models which should technically pass from a training experience to a next conserving the AI architecture in primis and thus also the data semantics, the format is less relevant technical aspects. In this sense, Snap4City with its Knowledge Base ontology, Km4City, can be a key to make any data interoperable at semantic level.
DATA MODELING AND INTEROPERABILITY
The world of Internet of Things, IoT, and its integration with Big Data are mature regarding the data collections, in the sense that any kind of data can be collected and aggregated into the most powerful platforms for smart city, Industry 4.0, health, energy, environment, etc. On the other hand, for most of those domains the actual needs are much wider. Snap4City is one of the most powerful platforms in terms of interoperability with high level types, FIWARE Smart Data Models, IoT Device Models, GIS data, satellite data, Origin Destination matrices, trajectories, shapes, traffic flow, flows in 3D, 3D Digital representation for cities, BIM, etc.
Snap4City includes Km4City ontological and semantic model (https://www.km4city.org) to guarantee the data interoperability with FIWARE Smart Data Models, IoT Data Models, and a large range of High Level Types, providing a number of real-time open source solutions to support decision makers in cities and large industries to ground their daily operational actions on solid ethical and explainable artificial intelligent predictions, deductions and assessments. It provides a complete understanding of the context and its trends, receiving early warning, anomaly detections, and performing simulations, early warnings and what-if analysis. This information is used to suggest strategic and real time interventions to improve city services and general quality of life, in multiple domains (e.g., mobility, transport, energy, government, tourism, environment, civil engineering).
In this large range of solutions, Snap4City is compliant and interoperable with more than 150 protocols fully integrating legacy systems, and it is interoperable with any GIS (Geographical Information Systems), BIM (local Digital Twin), CKAN (open data networks), Satellite, and IoT Networks protocols (IoT protocols), services and databases. https://www.snap4city.org/283, https://www.snap4city.org/65
The push on AI has been recently associated to the Digital Twin. Digital Twin aims are creating a digital counterpart of the physical entities and lead them working together. On the other hand, most of the physical solutions are not smart at all, this means that the Digital Twin will become the intelligent engine of the physical one in most of those cases.
READY TO USE DATA ANALYTICS
Snap4City means: Smart aNalytic APp builder for sentient Cities and IOT.
Snap4City has been designed since the 2017 to be ML/AI enabled, resecting ethics, secure passing the PENtest and GDPR compliant. Snap4City has developed a large number of solutions in the context of Smart City and Industry 4.0. Snap4City fully supports the development of real time data analytic processes through ML, AI, ethic trustworthy XAI via languages such as Python, R-Studio, also exploiting Tensor Flow, Pandas, Keras, and any kind of library for data analytics, ML and DL. Snap4City is distributing a number of Open-Source data analytics tools and algorithms for: prediction, anomaly detection, classification, detection, constrained routing, optimisation, analysis of demand vs offer of transportation, and many others have been published on international top level journals. Data Analytics is fully integrated into What-IF analysis tools in control rooms and for operators, defining scenarios and solutions. Snap4City Data Analytic course https://www.snap4city.org/download/video/course/da/ Snap4City has a consolidated experience in the development, validation and transfer AI/XAI solutions. Most of the DISIT lab solutions are based on ML, DL, AI, XAI, natural language processing (NLP), sentiment analysis (SA), semantic reasoning and computing. In the following, a number of examples are just listed, while more details can be recovered from the above mentioned Snap4City Data Analytic course and from technical notes: https://www.snap4city.org/4
Mobility and Transport
- Impact on planned changes in city viability for: public transport, traffic, parking, people flow, etc.
- What if analysis: routing, traffic flow, demand vs offer, pollutant, etc. (Simulation + ML)
- Traffic flow reconstruction from sensors and other sources (simulation + ML)
- Predictions for: traffic flow, smart parking, smart bike sharing, people flow, etc. (ML, DL)
- Public Transportation: Ingestion and modelling of GTFS and Transmodel
- Analysis of the demand mobility vs offer transport of according to public transportation and multiple data sources (Simulation)
- Assessing quality of public transportation (analysis)
- Accidents heatmaps, anomaly detection (analysis, ML)
- Tracking vehicle fleets, people, via devices: OBU, OBD2, mobile apps, etc.
- Routing and multimodal routing (multistop travel planning), constrained routing, dynamic routing
- Computing Origin Destination Matrices from different kind of data (analysis)
- Computing typical trajectories based on tracks (analysis, ML)
- Computing Messages for Connected drive
- Slow and Fast Mobility 15 Minute City Indexes (analysis, ML)
- Computing and comparing traffic flow on devices and at the city border (analysis)
- Typical time trends for traffic flow and IoT Time series (analysis, ML)
- Impact of COVID-19 on mobility and transport
City Users and Social
- People detection and classification: persona, carts, bikes, etc. (ML, DL)
- people counting and tracking, trajectories (via thermal cameras, ML, DL) in squares and locations
- People counting via head counting (via thermal cameras, ML, DL)
- People flows and counting prediction and reconstruction (ML, DL)
- Wi-Fi data, mobile apps data, Mobile Data, etc.
- Computing User engagement and suggestions for sustainable mobility (Rule Based, ML)
- User’s behaviour analysis,
- origin destination matrices, hot places, time schedule, Recency and frequency, permanence, typical trajectory, etc.
- People flow analysis from PAX Counters and heterogenous data sources
- Social media analysis on specific channel, specific keywords: see Twitter Vigilance,
- Reputation, service assessment: MultiLingual NLP and Sentiment Analysis, SA
- Tweet proneness, retweet-ability of tweets, impact guessing
- Audience predictions on TV channels and physical events, locations
- 15 Minute City Index, etc. (modelling and computability)
Environment and Weather
- Predictions of NOX, PM10 pollution on the basis of traffic flow, 48 hours
- Long term predictions of European Commission KPIs on
- NO2 average value over the year
- PM10, PM2.5, CO2, ...
- Prediction of landslides, 24 hours in advance (DL, XAI)
- Computation of CO2 on the basis of traffic flows, each road for each time slot of the day
- Prediction of Micro-Climate conditions for diffusion of: NO2, PM10, PM2.5, etc.
- Reduction for Optimisation of waste collection in large cities.
- Heatmaps production, dense data interpolation for
- Weather conditions: temperature, humidity, wind, DEW
- Pollutants and Aerosol: NO, NO2, CO2, PM10, PM2.5, etc.
- Impact of COVID-19 on Environmental aspects
Management and strategies
- What-if analysis, dynamic routing, origin destination matrices production from a large range of sources
- Early warning, anomaly computation
- Estimation of KPI and local indexes for: quality of life (15MinCityIndex, SDG, ..)
- Production Optimization
- Planning and Monitoring renovation works via objective KPIs
- Managing Maintenance and teams
- Predictive Maintenance and costs predictions: chemical plant, vehicles, boats (XAI)
- Resilience analysis wrt European Guidelines on Resilience of critical infrastructure, and
- Risk analysis: natural and non-natural disaster (XAI)
Semantic Reasoning
- Ontology Modelling and integration, expert system construction
- Multilingual Sentiment Analysis, transformers, BERT, Multilingual
- Knowledge modelling and reasoning on RDF stores: spatial, temporal, relational
- Virtual Assistant construction, virtual expert of the city.
Time Series, Matrices, Images, Maps and 3D Digital Models
- Time Series Anomaly detection
- Data quality assessment and control
- Short, long and very long term predictions
- Interpolation of scattered Data on regular grid for calibrated heatmaps
- Conversion of Satellite data images into regular ground images/heatmaps measures
- Extraction information from Orthomaps, LIDAR, etc., regarding city structures
- 3D Digital Twin of Cities and Objects: pattern extraction, 3D model reconstruction.
In 2019, DISIT Lab (University of Florence) turned out to be the winner of the Select4Cities PCP of EU and one year later won the ENEL-X open data challenge. Currently, Snap4City is one of the platforms of the EOSC (European Open Science Cloud), library of Node-RED, and DISIT Lab is proud to be a Gold Member of FIWARE and an official FIWARE Platform and Solution, certified Consultant, certified Trainer, and provides two certified FIWARE Experts. DISIT Lab and other partners participated providing Snap4City solutions and a strong number of innovations in a number of EC projects (HERIT-DATA, RESOLUTE, REPLICATE, TRAFAIR, MOBIMART, Select4Cities, Snap4City, WEEE, Panacea, Impetus, etc.), and national/regional (Sii-Mobility, MOSAIC, ALMAFLUIDA, SODA, Pretto, Enterprise, etc.), and in many direct contracts. Contact the coordinator of Snap4City to find out an idea of new technologies and solutions that will arrive in next 6/12 months.
Paolo Nesi
Prof. Eng. PhD Paolo Nesi is a professor at the University of Florence, UNIFI, working at DINFO, Department of Information Engineering (Computer Science Department), Chair of DISIT Lab of UNIFI, and chair of https://www. Snap4City.org which is an official platform and solution of FIWARE and of EOSC, Library on Node-RED, etc. P. Nesi main research topics are: big data analytics, AI/XAI, distributed systems, IoT, Cloud, security and privacy. He has more than 30 years of experience in developing predictive and prescriptive solutions based on semantic computing, machine learning, artificial intelligence, explainable AI, natural language processing, computer vision, etc., in a large range of domains, more than 400 international papers. He founded and is at responsible of the DISIT LAB (https://www.disit.org ). He has been scientific coordinator of large international research and innovation projects such as: Snap4City H2020 EC, Resolute H2020, Sii-Mobility national mobility and transport action in Italy, AXMEDIS EC, ECLAP EC, etc., and smart solutions responsible of many others such as Replicate H2020 EC, Trafair CEF, ICARO Cloud, HeritData EC, Mobimart EC, Weee Life, etc. He has been chair of IEEE SC2, IEEE ICSM, IEEE ICECCS, AXMEDIS, DMS, .. and programme chair of many others. He is editor of international journals, member of the national Italian center of mobility, member of the CBDAI (Center for Big Data and Artificial Intelligence of the Tuscany Region), member of the EDIH Tuscany-X board, scientific board member of the PhD-AI National PhD Course on Artificial Intelligence, certified FIWARE Expert, scientific board member PhD course of UNIFI on Information Engineering, advisory board member of a number of EC International Projects. He has been the recipient of a number of Awards as: recognized top 15 researchers worldwide in the area of software engineering for two years, Snap4City First place awards, best paper awards, etc., and It has been co-chair of SMR MPEG-4. P. Nesi is a member of IEEE, ACM, AI*IA, CINI, CNIT, ISO, FIWARE, Gaia-X.
Google Scholar: https://scholar.google.com/citations?user=c2S3Ni0AAAAJ&hl=en
Twitter: https://twitter.com/paolonesi
Facebook: https://www.facebook.com/paolo.nesi2
LinkedIn: https://it.linkedin.com/pub/paolo-nesi/1/ba5/849
YouTube: https://www.youtube.com/channel/UC3tAO09EbNba8f2-u4vandg
