Understanding image and languagewith deep learning for e-commerce

Charlotte Boutelier

Laurent Ach (European Director of Rakuten Institute of Technology, the research department of Rakuten group) will be presenting on September 15th, and will share his knowledge through a keynote speech entitled “Understanding Image and Language with Deep Learning for E-Commerce”. 

What are the main areas of research of the Rakuten Institute of Technology? What is its overall mission?

The mission of Rakuten Institute of Technology is to do applied research especially in artificial intelligence and to leverage scientific results for the benefit of Rakuten business and customer experience. Our research focuses on three main areas: customer behavior analysis and prediction, natural language processing, and computer vision. Messages from customers, description of products, are made of natural language, and dealing with images is essential in e-commerce.

According to you, which sectors will benefit the most from AI? Why?

Some tools based on machine learning are broadly used since the beginning of e-commerce, like in recommender systems. Several machine learning techniques, including neural networks but also a bunch of other statistical methods are now deployed for marketing applications, fraud detections, credit scoring, and for the prediction or optimization of various types of data. Applications of deep learning achieve amazing performances and gain visibility for the general public, like face recognition, speech recognition, language translation, or computers playing games.

Then, there are domains where AI is used but currently reaches technical limits or lack reliability, like autonomous driving, medical applications, virtual assistants. In other domains, AI just needs more adoption and integration and there is potential for creating tools automating or helping humans with many tasks. Internet companies are the first to benefit from AI but gradually all sectors will be impacted.

What other technologies are currently being worked on at the RIT? Which one is the most promising, according to you?

Deep learning in various forms is now at the basis of many AI projects but it cannot be applied blindly, we need to analyze the biases in data, find causal relationships between events, explain some predictions, and these tasks are not intrinsically part of deep learning approach, which currently needs huge amount of data for training, lacks common sense reasoning, and is mostly a black box. So, we also need other existing AI techniques, which we use at RIT. Still, Deep Learning is what allows us to achieve an unparalleled level of automation of tasks that where recently only achievable by humans, especially when it comes to dealing with images and natural language.

Share this: