Abstract: Deep Learning (DL) has demonstrated undeniable successes in object detection, language translation, video games, etc. However, DL is simply classical machine learning and statistical methods powered by GPU acceleration. Deep Learning networks only compute what they have been programmed or trained earlier, fail in a way that humans will never do, and are not able to learn from errors. For example, the dangers represented by adversarial attacks create serious problems to self-driving vehicles or cybersecurity. They can be partially recovered by Bayesian DL to model uncertainty, but for the complex use cases we are confronted today, it is not enough since we need to build a Human-Level Artificial Intelligence (HLAI) that learns from few examples, is goal oriented, improves continuously from errors, provides explainability and understandability, and learns how-to-learn for acquiring a generalized knowledge to new situations. In this talk, I will present promising routes towards these goals based on some advanced cognitive architectures such as openNARS.
Abstract: Amazon SageMaker is a fully-managed platform that enables developers and data scientists to quickly and easily build, train, and deploy Machine Learning and Deep Learning models at any scale. In this session, we’ll introduce you to the service and we’ll run Python notebooks solving problems with both built-in algorithms (XGBoost, K-Means, etc.) and your own custom code (TensorFlow, Keras, etc.).
Abstract: The entertainment business in the area of Festivals has grown tremendously over the last years. This big business of (typically) young DJ’s, management agencies, and festival entrepreneurs is a serious business ecosystem with Events, Merchandise, Music Streaming, Sponsorships and royalties.
The competition for Fans and their engagement, content consumption and spending is huge. In this session, Edwin will share how their technology on Data and Intelligence provides differentiating value in strategies of Festival owners and Artists.
Abstract: omni:us provides AI-powered document processing services for the insurance industry in order to fuel the digital transformation and make existing workflows more transparent, affordable and efficient. Handwritten forms are an important part in a large variety of insurance use cases such as claims. Extracting information from forms is a well-defined but challenging task due to multiple form versions, varying scan and photo quality, handwriting from unknown writers and a large amount of entities per form. This talk will provide an overview of our processing chain including page classification, template alignment and handwritten text recognition and discuss our key findings from various real-world projects.
Abstract: We live in a time when information about most of our movements and actions is collected and stored in real time. The availability of large-scale mobile phone, credit card, browsing history, etc data dramatically increase our capacity to understand and potentially affect the behavior of individuals and collectives.
The use of this data, however, raise legitimate privacy concerns. In this talk, I will discuss how traditional data protection mechanisms fail to protect people’s privacy in the age of big data. More specifically, I will show how the mere absence of obvious identifiers such as name or phone number or the addition of noise are not enough to prevent re-identification and how sensitive information can often be inferred from seemingly innocuous data. I will discuss some of our recent work on privacy in networked environments, and the development of an attack on a commercial privacy-preserving software. I will then conclude by discussing some of socially positive uses of big data and solutions we are developing at Imperial College to allow large-scale behavioral data to be used while giving individual strong privacy guarantees.
Abstract: Realistic music generation is a challenging task. When machine learning is used to build generative models of music, typically high-level representations such as scores, piano rolls or MIDI sequences are used that abstract away the idiosyncrasies of a particular performance. But these nuances are very important for our perception of musicality and realism, so we embark on modelling music in the raw audio domain. I will discuss some of the advantages and disadvantages of this approach, and the challenges it entails.
Abstract: Increasing pressure of regulation authorities and the need of risk mitigation broaden the space for applications of artificial intelligence even in the conservative environment of banks. Traditionally, banks have been employing simple methods and spending enormous resources to run their customers through anti-money laundering screenings. Using machine learning and natural language processing, we can improve screenings in both time efficiency and quality… helping the fight against evil villains. The talk presents how our ML models identify (not only) criminals, tells the story of our deep learning models and shows the challenges that are in front of us.
Abstract: will look into the growing challenges faced when programming heterogeneous computers and look forward to the solutions that developers can use in the future.