Pr. Boualem Benatallah
University of New South Wales Sydney
Pr. Boualem Benatallah received his PhD degree in computer science from Grenoble University (IMAG, France). He is Professor at the University of New South Wales Sydney, Australia. His research interests lie in the areas of Web service protocols analysis and management, enterprise services integration, large scale and autonomous data sharing, process modeling and service oriented architectures for pervasive computing. He has several ARC (Australian Research Council) funded projects in these areas. He was a visiting scholar at Purdue University (USA), Visiting Professor at INRIA-LORIA (France), Visiting Professor at Blaise Pascal University (Clermont-Ferrand, France). He has been a Program Committee member of several conferences. He was the PC chair of several Int. workshops on Web services. He is the PC co-chair of the 3rd Int. Conf. On business Process Management (Nancy, France, September , 2005) and 3rd Int. Conference on Service Oriented Computing (December 2005, Amsterdam). He was guest editor of several journal special issues on Web services. He was keynote and tutorial speaker on Web Services at several workshops and conferences. He is co-author of Interconnecting Heterogeneous Information Systems (Kluwer,1998). He is also co-author of E-Commerce Enabling Technologies, Pearson Education, 2002. He has published widely in international journals and con-ferences including IEEE TKDE, IEEE TSE, IEEE Internet Computing, IEEE Net-work, IEEE Intelligent Systems, VLDB, PADD journals and IEEE ICDE, IEEE ICDS, WWW, ER conferences.
Talk 1: From APIs to Conversational Cognitive Services
Cognitive services and their instantiation in the form of messaging or chat bots, task-oriented conversational bots, software robots, digital or virtual assistants, are today used by millions of users. Increasingly, organisations have started to use conversational AI to augment and improve productivity and effectiveness of their customers, workers and stake-holders, automate business processes, deliver data-driven insights. However, there are significant gaps in the cognitive service-enabled endeavour. From engineering perspective, bot developer defines goals for the bot, selects or creates relevant user intents and API(s) providing the service, trains the bot extensively to identify the intents and parameters, and manually translates intents into API calls and conversations with users. We postulate that the ubiquity of cognitive services will have little value if they cannot easily integrate and reuse concomitant capabilities across large number of evolving and heterogeneous devices, data sources and applications. At the same time, APIs are unlocking application, data source and device silos through standardised interaction protocols and access interfaces. Today, within the Web and Mobile development community, complex applications are being stringed together with a few lines of code – all made possible by APIs and their composition. To leverage the opportunities that APIs bring, we need bot development to 'scale' in terms of how efficiently and effectively they can integrate with potentially large number of evolving APIs. We will discuss some critical challenges to achieve this objective. First, a core challenge is the lack of latent and rich intent and APIs knowledge to effectively and efficiently support dynamic mapping of complex and context-specific user intents to API calls. Second, user intent may be complex and its realisation requires composition of multiple APIs (e.g., triggering multiple APIs to control IoT devices using one user utterance). Existing intent composition techniques typically rely on inflexible and costly methods including extensive intent training or development of complex and hard-coded intent recognition rules. We will discuss challenges in API aware training of cognitive services. We will discuss novel latent knowledge-powered middleware techniques and services to accelerate bot development pipelines by: (i) devising novel intent and API element embeddings and matching techniques, (ii) declaratively specifying reusable and configurable conversation models to support complex user intent provisioning; and (iii) dynamically synthesising API calls instead of ad hoc, rule-based and costly development of intent-to-executable-code mappings.