October 24, 2022
The platform, designed for debugging neural networks, allows data scientists to understand how the model perceives the data, makes decisions, in which cases it fails, and why. The company completed a seed round of $5.2 million
The startup company Tensorleap exposed with a platform For debugging and deep testing of artificial neural networks. According to the company, the system drastically reduces the danger that the models developed by the data scientists will fail after the transition to the real world (production environment).
The company was founded by Dodi Ben David (CEO), Yotam Azriel (VP of Technology) and Nir Ben David (CCO). The company raised a seed of 5.2 million dollars from the funds Angular Ventures, Sozo Ventures, Industry Ventures and employs 15 people in the development center in Ramat Gan.
Widespread adoption of neural networks began less than a decade ago, with the availability of high computing power, large volumes of data and the ability to efficiently and quickly access the cloud. Neural networks address problems and applications that classical machine learning models cannot deal with, in areas such as industrial production and medical technologies to autonomous vehicles, finance and cyber. Therefore, it is estimated that the use of neural networks will expand exponentially as artificial intelligence reaches new fields and uses.
deep neural networks (DNN) are conducted in a hidden manner in a kind of “black box”, which means that the development cycles are mainly based on trial and error. Therefore, they are long, ineffective and expensive, and worse, their results are unreliable. Even after achieving the desired accuracy, data scientists must verify the robustness of the models in extreme cases, to avoid failures in these scenarios in production. Even in cases where models function as expected, the data scientists have no ability to understand and explain how the model reached the result. When it comes to life-affecting decisions, such as medical scan diagnostics or autonomous driving, the ability to verify and explain success and failure are critical to mission success.
The existing methods for developing neural networks are mainly based on experiment management systems that help data scientists try more and more, consuming a lot of development time and enormous computing power. With the help of advanced algorithms developed by the company, Tensorleap Helps to open the “black box”, so that data scientists can understand how the model perceives the data, how it makes decisions, in which cases it fails, and most importantly – why. With this visibility, data scientists can identify and solve problems, analyze underrepresented populations, address overfitting (suspicious overfitting of the data), perform comprehensive tests for the model and make informed decisions about which models are really ready to be implemented in production.
Tensorleap is a subscription-based platform that works in a model SaaS or installed on the client’s website or cloud and supports all types of data, including images, text, graphs, tabular information and more. The company has clients in a variety of fields, from medical products to chip manufacturers, gaming and social media platforms.
[קרדיט צילום: אורן דאי]
Posted in the categories: big data, artificial intelligence, capital raising, news
Posted in tags: Tensorleap , neural networks
