Mice teaching AI, Turning CO2 into Fuel and Your Breath is like a Fingerprint

June 19

This week we discover how mice can teach AI a thing or two. We investigate a way that battery free RFID chips can provide information about the world around us. We look at a new vastly improved method for turning CO2 into fuel and other useable chemicals. Finally we examine a way to use our breath like a fingerprint for identification and to investigate a host of other issues.

Mice teaching AI

This is not a comment on the current capability of AI, rather it is about a better way of learning. A team at Technical University of Munich have discovered that Artificial Neural Networks can predict movements more effectively when trained on biological data from early visual system development. Movement prediction is important for self driving cars, robotics and other real world applications of AI.

Before our eyes open, built in training begins in the retina. This training is independent of external stimuli and is also true for mice, cats and dogs. Spontaneous activity patters spread in wave like motions across the eye’s neural tissue. These are known as retinal waves. They build the early wiring between the brain’s visual system and the retina.

AI based neural networks are typically trained using data that resembles the task that they are designed to perform. In an animal world, this would represent visual learning only beginning when they open their eyes whereas we know it begins before the eyes are opened.

The team pre-trained a neural network with retinal wave data from a mouse. The network was then trained using an animated film simulating the perspective of a mouse running through a narrow corridor lined with various patterns. A second neural network was only trained using the animated film, i.e. no pre-training.

Each network has to accurately predict how the patterns on the simulated walls would evolve. The networks pre-trained with retinal waves performed the task more accurately and more quickly than those without pre-training.

A further test was carried out using a grainy film of real world footage from a cat’s perspective i.e. showing what a cat sees. The video was lower quality and the movement more complex, however the pre-trained neural networks once again outperformed all others.

Battery Free RFID sensing of Real World Data

A team from University of California San Diego and Qualcomm have developed a real time RFID based passive sensing system. The system can measure naturally occurring phenomena using harvested radio frequency energy. Phenomena such as temperature, pressure or weight.

The team used a differential sensing system that uses two RFID integrated circuits connected to the same antenna to interfere with each other. Instead of data being sent simultaneously, data would arrive sequentially. An algorithm was used to unite the two divergent sequences and make some reliable sensor data without wires or batteries.

There are about 50 billion RFID devices used each year and they are connected to 4 to 5 billion bluetooth devices. Commercial RFID tags follow strict communications protocols that include frequency hopping or sequential tag reading. This leads to distorted signals and timing mismatches.

The team used an algorithm called Dynamic Time Warping (DTW) that was originally used in speech recognition. The algorithm uses differential signal behaviors to get more reliable readings. The system can process up to 500 data samples per second with sub-degree error rates in complex, dynamic environments.

The system may be able to help automate warehouses, monitor agriculture and improve medical sensing and reduce food waste. A smart phone pointed at a box in a warehouse could measure its’ weight. The moisture content of a pot of soil could be visualized. The system can operate indoors, outdoors and in mobile environments.

Turning CO2 into Fuel

Reusing CO2 that has been released by burning fossil fuels does not reduce the amount of CO2 in the atmosphere but it does stop adding more CO2 into the atmosphere. It is known as cyclical carbon and it can be a useful step towards reducing overall carbon emissions.

A team at Rice University have developed a simple way to improve the conversion of CO2 into fuel and chemicals. Current systems suffer from a build up of salt that clogs the gas flow channels. This reduces efficiency and causes premature failure of the systems.

Electrochemical CO2 reduction uses electricity (preferably from renewable sources) to turn co2 into valuable products like carbon monoxide, ethylene or alcohols. These products can be further refined into fuels and other feedstocks for industrial processes.

A technique called acid-humidified CO2 was used to improve the operational life of a CO2 reduction system 50 fold. The team used water to humidify the CO2 gas input into the reactor by bubbling the gas through an acid solution. The vapor from the gas carries into the reaction chamber in large enough quantity to affect the chemistry of the reaction. This stopped the salts from blocking the gas channels.

The result was a device that operated for over 4,500 hours verses 80 hours previously. The method works with different catalysts (which generate different outputs) and is easily scalable without affecting energy efficiency.

Your Breath is like a Fingerprint

A team from the Weizmann Institute of Science in Rehovot Israel has developed a way to identify humans by their unique breathing patterns. Further the team found correlations between breathing pattern and BMI, measures of anxiety, depression and autism.

Participants had their breathing measured over a 24 hour period by a wearable lightweight portable device with nasal tubes and motion sensors. They recorded their daily activities in a phone app. The device continuously recorded airflow passing through the nasal passage.

The team extracted 24 parameters from the participants airflow. The data was separated into asleep and awake pools. A machine learning algorithm was then trained on the data. Approximately half of the participants returned to the lab to be remeasured months and in some cases two years later. The algorithm was able to identify the individuals from their breathing patters.

The current device has its’ limitations as it can move during sleep and is awkward in appearance (tubes are placed in the nose) making it less suitable for daily use. The team is working on a more compact user friendly device that will allow a larger and longer study.

Paying it Forward

If you have a start-up or know of a start-up that has a product ready for market please let me know. I would be happy to have a look and feature the startup in this newsletter. Also if any startups need introductions please get in touch and I will help where I can.

If you have any questions or comments please comment below.

I would also appreciate it if you could forward this newsletter to anyone that you think might be interested or provide a recommendation on Substack.