KAIST Develops Semiconductor Neuron That Tunes Noise to Selectively Process Signals (IMAGE)
Caption
Figure 3. Reconfigurable time-series encoding for human activity and speech signals
By adjusting the memristor resistance state, the encoding characteristics can be tailored to low-frequency human activity signals from the UCI HAR dataset or high-frequency speech signals from the Audio MNIST dataset. The original time-series signals are converted into probability maps that provide feature information for subsequent AI classification. Using these encoded features, the system achieved approximately 95% classification performance for both human activity and speech recognition.
Credit
KAIST
Usage Restrictions
No
License
Licensed content