Alma Mundi Ventures Invests In AI Startup Nnaisense, The Pioneers Of Very Deep Learning

Alma Mundi Ventures is a venture capital firm focused on technology-based companies, which emphasizes business to business models that are raising their series A or B rounds. They invest between $500,000 and $5 million in each company through their first fund.

LUGANO, Switzerland, MADRID and SEATTLE, Jan. 17, 2017 /PRNewswire/ -- Today, Alma Mundi Ventures announced a Series A investment in Nnaisense, an artificial general intelligence (AGI) and very deep learning start-up led by pioneers in the field. AGI is the intelligence of a machine that can perform any intellectual task at a human being level.

Alma Mundi Ventures is a venture capital firm focused on technology-based companies, which emphasizes business to business models that are raising their series A or B rounds. They invest between $500,000 and $5 million in each company through their first fund.

Nnaisense's president and co-founder is Jürgen Schmidhuber, scientific director at the Swiss AI lab, IDSIA and luminary in the fields of reinforcement learning and neural networks, especially recurrent neural networks (RNNs) which allow AI machines to learn to remember.

In 1997, Schmidhuber and Nnaisense Advisor Sepp Hochreiter co-authored the seminal paper1 about the most powerful RNN, Long Short-Term Memory (LSTM). LSTM is used by Amazon, Apple, Baidu, IBM, Google and Microsoft to solve real-world problems today. Unlike traditional RNNs, LSTM can efficiently learn from experience to classify, process and predict time series when there are long time lags between key events. Schmidhuber is known as "the guy who taught AI to remember."

The other co-founders, all former IDSIA researchers, are Faustino Gomez (CEO), Jan Koutník, Jonathan Masci, and Bas Steunebrink, are pioneers of very deep learning and reinforcement learning due to several technology firsts.

An RNN, like the human brain, is a network of neurons with feedback connections that can learn many behaviors not achievable by traditional machine learning methods. Artificial RNNs can learn algorithms that map inputs to outputs with or without a teacher and therefore are computationally more powerful and biologically more plausible than any other current adaptive approach.

Recent applications of RNNs include adaptive robots, handwriting recognition, keyword spotting, music composition, attentive vision, protein analysis, machine translation, speech recognition and stock market prediction.

"Alma Mundi Ventures aspires to invest in global technology innovation leaders. Nnaisense is at the forefront of next generation AI," said Rajeev Singh-Molares, founding partner, Alma Mundi Ventures. "And Nnaisense is off to an impressive start with high profile projects with Audi and Acatis amongst others," he added.

The Alma Mundi investment will be used to hire top engineers and researchers to satisfy the growing need for Nnaisense's technology in industries such as manufacturing and financial services. Acatis Investment GmbH, which manages over €3 billion, is already using Nnaisense AI programs for financial prediction for asset management.


Nnaisense scientists, through their IDSIA work, have several very deep learning "firsts" and technology records. Four "firsts" stand out.

In May of 2015, Nnaisense scientists invented Highway networks, the first deep networks capable of being trained with over 100 layers. Going deeper enables networks to solve more complex real-world problems. Very deep networks can allow machines to better understand context, for example, in a driving or conversation scenarios, and infer answers to complex questions.

They also hold the record for deepest recurrent networks. Note that unlike convolutional neural networks which focus solely on solving static problems like pattern recognition, recurrent networks work on more general, dynamic or time-sensitive problems like speech, translation, and long-term investment predictions. In 2016, Nnaisense scientists developed successful recurrent Highway networks (RHNs) that utilize tens of layers at each time-step of the process. These networks are much more effective than previous ones which could only use two or three layers per time-step.

Nnaisense is engaged in a variety of industrial applications of deep learning including defect detection and material classification which are currently being tested by a large steel producer. But the mission of the company goes beyond artificial perception to close the sensory-motor loop. It does this by building systems that learn by themselves to control a variety of complex environments, such as chemical plants and autonomous robots, through reinforcement learning. In 2013, they became the first to successfully train recurrent neural networks to drive a simulated car using vision, without the aid of a teacher.


On stage at the conference and workshop on Neural Information Processing Systems (NIPS) last month Nnaisense technology was used to autonomously park an Audi model car. This demonstration was made possible by close collaboration between Audi AG and Nnaisense in developing a deep reinforcement learning system that trains the car to park in simulation, using only raw camera data. Once trained, the resulting recurrent neural network "brain" is transferred to the car where it smoothly controls steering, accelerator, and brakes, to maneuver it to the parking spot from any start position. This was the first time reinforcement learning was used to park a vehicle; the significance of this milestone is that the model car learned without a teacher.


Two things make Alma Mundi Ventures unique: its "Spanish diaspora" investment strategy which aims to invest in promising technology companies all over the world and its portfolio companies' access to its Mundi Club. The Club is made up of senior executives, including many CEOs, from hundreds of companies including Amazon, Bertelsmann, Cisco, and Intel to name a few. The club gets together in an annual forum and quarterly in different locations with the startups that it supports. Its network is used not only to conduct due diligence but also to accelerate their growth, by facilitating access to corporate executives, new customers and new markets.

Featured Product

Pleora Technologies – Protect Your Brand with  Visual Inspection

Pleora Technologies - Protect Your Brand with Visual Inspection

A unique distillery uses Pleora's Visual Inspection System to maintain brand quality standards and reduce errors that increase operating costs and slow production. To help ensure brand consistency, the distillery uses Pleora's AI-based Visual Inspection System to add decision-support for manual labelling processes. The camera-based system integrates pre-packaged AI plug-ins that are easily trained on a customer's unique requirements to visually highlight product differences and deviations for operators and inspectors on a display. The distillery trained the Image Compare plug-in to identify key brand elements on the bottle, with a real-time on-screen image overlay then guiding operators as they manually place the emblem. In addition, the system can be used as a training tool to teach new employees on brand quality standards.