Artificial Intelligence for Business

AI is taking more and more importance for business. This conference exposes the relations between artificial intelligence, different industries and cybersecurity.

Artificial Intelligence for Business

AI is taking more and more importance for business. This conference exposes the relations between artificial intelligence, different industries and cybersecurity.

IA for the Travel Industry: A Smooth Journey from Inspiration to Payment

It’s sometimes difficult to understand why people travel, because of a lack of data. Airliners operate global networks and they don’t know much about passengers. They also have to identify new route opportunities. The company Milanamos helps them understand which capacities they could put in any route over the next 3 to 5 years using an AI algorithm which is predictive.

In Europe, multimodal or intermodal transports are a political debate. Milanamos collects data from everybody and on top of that, they get a lot of data coming from GPS or mobile phones. They map how they should dispatch and offer better connections between transport operators, using algorithm, AI and operation research. They recommend what should be the ideal connection. Today, 35% of travelers don’t use public transportation to get to the airport due to poor connectivity.

The key challenge is to reduce gas emission and the usage of personal cars to go to the airports. AI is used to recommend the ideal transport connection, setting different objectives in terms of optimization, such as environment or revenue optimization.

They use smart contract and AI to match all the tickets sold and to manage settlement of revenue between operators. This company is not BtoC but entirely BtoB. They help people create value-based prices and recommend what should be the ideal prices. In order to do that, they focus on what the consumer does and what he likes.

Speaker: Christophe IMBERT, Milanamos

Credit Card Fraud Detection Using Automated Machine Learning

The goal of MyDataModels is to give to security experts the possibility to build automatically predictive models for credit card fraud detection. Today, four main facts must be taken in consideration:

  • Decisions are increasingly data-driven.
  • The available data is more often small data than Big Data. This one is not the most frequent in business.
  • Domain experts in terms of security (industrial, researchers) have limited or no data science skills (coding, programming, machine learning algorithm). The way you work with a data scientist is a long process, time consuming, repetitive, and expensive. The way this company works is completely automated. In a few clicks, you get your data and you get your predictive models.
  • When available, data analysis is time consuming, complex and expensive.

Using automated machine learning, you get your data in the software, and then you try to find what the most important variables are. Your model is automatically built, and you use performance metrics to evaluate the efficiency of the process.

Speaker : Simon GAZIKIAN, MyDataModels

Practical Usage of AI in Cybersecurity

Global AI business has reached 1, 2 trillion in 2018. According to the BBC, 800 million jobs might be taken by robot automation. But this market is growing. Companies start to see the interest in cybersecurity.

But there are still a lot of incidents and data breaches. In 2016, cybercrime caused over 100 billion lost to the US economy. The UK confirms that 7 out of 10 large companies identified breaches last year. There are disparities between AI and cybersecurity.

AI covers different meanings and is often misused. AI should be able to learn and progress in terms of skills and capacity to solve tasks. Machine learning can outperform human because of a bigger capacity to recognize images. But it can’t replace human beings, because even if it could replace them for repetitive tasks, the continuous cost to maintain this technology is huge.

In terms of cybersecurity, there’s a lot of talented people, but no qualified people to implement risks based, consistent and coherent cybersecurity program in a company:

  • Near 92% of external web application have exploitable vulnerabilities
  • 19% of the companies have external unprotected cloud storage
  • 2% only of external web applications are protected with a web application firewall.

AI will not replace a cybersecurity strategy, but it can support, guide and provide some meaningful numbers. Strong AI, fully capable of replacing us, does not and will probably not exist within the next decade. It might create new jobs and provide great support, reducing time spending in trivial routine tasks.

Although, it won’t repair fundamental things, for example if you don’t have a risk based, comprehensive and holistic cybersecurity program in your company. In that case, don’t invest in AI because you are going to lose money.

Speaker: Ilia KOLOCHENKO, High-Tech Bridge

Intellectual Property protection for Distributed Neural Networks: Data, Model and Inference Confidentiality Preserving

Today, if we look at ERP systems, the unique source of truth about access points is cloud. The company SAP Labs France is looking forward to moving their ERP systems to the edge, as close as possible to the business. They are also trying to expand their security frameworks from 3 perspectives:

  • Data protection
  • Device
  • Application

If we push those three perspectives, we have to protect them. It’s important to leverage on machine learning technologies to do predictive maintenance, to get some information from video systems, extracting some insights out of the video stream to do some risk prevention on public spaces. The objective there is to improve the operational efficiency of those solutions.

But there are some problems in terms of security. Some inputs (data) are collected and injected from distributed systems (video streams, information from machines). They process that data to create high value inside it. Video stream is useless if you don’t automatically extract information out. Three major key resources are required: data, infrastructure and knowledge.

As part of the value chain of the machine learning, this company has to guarantee the compliance between input, output and the international data regulation (GDPR). They have to protect the machine learning algorithms which are important assets, and to do that, they use fully homomorphic protection:

  • First of all, you train your model base on knowledge. It takes time and costs (infrastructure and a lot of data).
  • You protect that, encrypting your model. It’s based on a PKI approach.
  • You push your encrypted model together with a key.
  • When you receive some data, you can encrypt it with that key, inject it in your model and obtain an encrypted inference.
  • That inference is sent to the network and when it has reached the back-end, you can decrypt it.

With this process, no one in the middle can get access to the model and data confidentiality is also protected.

Speaker: Laurent GOMEZ, SAP Labs France