Facial Recognition to Beer Tasting: Akida Neuromorphic Processor in Action

What is Akida Edge?

The Akida neuromorphic processor from BrainChip is a third-generation neural network designed to accelerate advanced AI at the edge. It is created after researching the neuronal architecture of the human brain. The Akida processor aims to accelerate neural networks such as recurrent neural networks (RNNs), deep neural networks (DNNs), convolutional neural networks (CNNs), and vision transformers (ViTs) in hardware.

It also supports Temporal Event-based Neural Nets (TENNs), a unique neural network architecture designed for quick analysis of complex time-series data.

In their recently released video, BrainChip demonstrates how the Akida can be applied. 

Cutting Edge Applications of Akida Edge

1. Time-of-Flight (ToF) Anti-Spoofing

Time of Flight (ToF) anti-spoofing is a technology used to enhance security systems by accurately detecting and preventing spoofing attempts. One notable application of ToF anti-spoofing is in facial recognition systems, where the technology can differentiate between a real person’s face and a spoofed image or mask to ensure secure access control. 

This feature is demonstrated in the BrainChip demo reel by showcasing Akida Edge learning with ToF sensors. The system accurately identifies the man named Rob. However, when Rob attempts to deceive the system by holding an image in front of his face, the system initially categorizes it as 'unknown.' This demonstration illustrates how Akida can discern between a genuine face and a spoofed image or mask.

2. Keyword Spotting

Keyword spotting is a speech recognition approach for identifying individual words or phrases in audio. Keywords are provided to the system that trigger an action. For example, a security system may be trained to listen to the sound of breaking glass and set off an alarm upon detection. 

In the BrainChip demo reel, this feature is demonstrated by a drone. When the man says 'on,' the drone begins functioning. It ascends with the sound 'up,' and descends with the keyword 'off.' 

3. Vibration Analysis 

Vibration Analysis is a diagnostic technique for detecting, tracking, and preventing malfunctions in revolving and rotating machinery. It involves the assessment of vibration patterns to identify deviations indicative of potential issues. 

For instance, it is employed in monitoring the performance of turbines, motors, pumps, and other rotating equipment in various sectors, including manufacturing, energy, and transportation.

In the BrainChip demo reel, this feature is demonstrated by a fan. When the fan is activated, Akida accurately predicts the normal vibration pattern as ‘on.’ However, when an external factor, such as inserting a paper into the fan, causes vibrations to become distorted, Akida promptly detects this anomaly and predicts a defect. 

4. Facial Recognition

Facial recognition is a biometric technology used to identify and verify individuals by analyzing facial features. It is used across various sectors, including security, access control, and personalized user experiences in consumer electronics.

In the BrainChip demo reel, a recognition system powered by the iToF AF0130 Devkit + Akida IP, utilizing indirect Time-of-Flight (iToF) technology, is showcased. 

Initially identifying individuals as "Unknown," the system quickly learns to recognize them upon inputting their names. Notably, individuals named Nolan and Chris are accurately identified after their names are entered. This demonstration highlights the system's adaptability and learning capabilities.

5. Beer Tasting with Akida

Beer tasting allows you to discern the origins, composition, style, ingredients, and overall characteristics of specific drinks. Beer tasting is crucial in brewery quality control, consumer taste testing, and beverage industry research and development. 

In the BrainChip demo reel, different beers are ‘tasted’ using Akida. Each beer is analyzed by Akida, which produces classification results based on the predominant taste ingredient. Initially, a sample is tested, revealing pale ale as the top taste, followed by other classifications. 

Subsequently, another sample is tested, with stout identified as the winning taste, prompting Akida to announce, 'I tasted Stout.' This showcases Akida's ability to accurately classify beer tastes.

6. In-cabin Visual Wake

In-cabin visual wake is a sensor-based technology integrated with Akida AI that detects human presence and identifies individuals entering or exiting a cabin or enclosed space. You may use the in-cabin feature to make your home ‘smart’ by providing proactive security alerts, intuitive voice commands for smart device management, and detecting the presence of any approaching person for several tasks.

In the BrainChip demo reel, the in-cabin visual wake option powered by Akida AI sensor technology is showcased. Using an automobile as an example, the system demonstrates its ability to detect human presence and recognize individuals entering or exiting the car. For instance, when Aaron enters the car, the system identifies his presence. 

Furthermore, the system demonstrates simultaneous functions:

• Akida Keyword Spotting: Aaron approaches and triggers the system by saying "Hey Akida," to which Akida responds.

• Akida Face Recognition: Akida recognized the individual and said, "Hey Aaron," just as he sat in the car.

• Akida Visual Wake Words: Following Aaron's interaction, when Tod opens the car door, Akida detects his presence.

7. Regression Analysis

Regression Analysis is a statistical technique used to model the relationship between a dependent variable and one or more independent variables. It can be used in civil engineering and infrastructure maintenance to predict structural issues in buildings, bridges, etc. 

In the BrainChip demo reel, Akida is showcased for regression analysis to monitor the structural health of a bridge. The demonstration illustrates how Akida predicts regression as ‘normal’ during routine operations. However, anomalies detected by Akida signal potential structural issues, prompting an ‘ALERT’ prediction.

8. Gesture Recognition

It is a user interface technology that uses cameras or sensors to capture and interpret human gestures, translating them into actionable data. It is used for intuitive interaction with digital devices and virtual objects.

In the BrainChip demo reel, gesture recognition is showcased as a compelling application of Akida technology. The video demonstrates how Akida interprets and responds to different hand movements captured by Time-of-Flight (ToF) sensors. For instance, the demonstration features a man performing various hand gestures, ranging from a hand clap to playing air guitar, which Akida associates with percentage values.

Watch the full video to understand better the applications of neuromorphic technology.