We present the Surrogate-gradient Online Error-triggered Learning (SOEL) system for online few-shot learning on neuromorphic processors. The SOEL uses a combination of transfer and principles computational neuroscience deep learning. show that partially trained Spiking Neural Networks (SNNs) implemented hardware can rapidly adapt to new classes data within domain. updates trigger when an error occurs, enabling faster with fewer updates. Using gesture recognition as case study, we be used...

Abstract Adaptive ‘life-long’ learning at the edge and during online task performance is an aspirational goal of artificial intelligence research. Neuromorphic hardware implementing spiking neural networks (SNNs) are particularly attractive in this regard, as their real-time, event-based, local computing paradigm makes them suitable for implementations fast learning. However, long iterative that characterizes state-of-the-art SNN training incompatible with physical nature real-time operation...

As voice assistant usage continues to grow, their homogeneity becomes even more problematic with the UNESCO report, "I'd Blush if I could" showing that designing only feminine assistants encourages negative behavior, both virtual and real people [3]. While masculine text-to-speech (TTS) voices exist, ones cover full range of gender presentations, such as non-binary or gender-ambiguous are largely missing. In this paper, we present a method creating TTS an example voice, Sam, created input...

Recent work suggests that synaptic plasticity dynamics in biological models of neurons and neuromorphic hardware are compatible with gradient-based learning [1]. Gradient-based requires iterating several times over a dataset, which is both time-consuming constrains the training samples to be independently identically distributed. This incompatible systems do not have boundaries between inference, such as hardware. One approach overcome these constraints transfer learning, where portion...

Despite the maturity and availability of speech recognition systems, there are few available spiking tasks that can be implemented with current neuromorphic systems. The methods used previously to generate data not capable encoding in real-time or very large modern datasets efficiently for input processors. ability encode audio spikes will enable a wider variety also algorithmic development automatic Therefore, we developed speech2spikes, simple efficient processing pipeline encodes recorded...

In this paper, we proposed a novel zero velocity detector, the Dynamic-Vision-Sensor (DVS)-aided Stance Phase Optimal dEtection (SHOE) for Zero-velocity-UPdaTe (ZUPT)-aided Inertial Navigation Systems (INS) augmented by foot-mounted event-based camera DVS128. We observed that firing rate of DVS consistently increased during swing phase and decreased stance in indoor walking experiments. experimentally determined optimal placement configuration zero-velocity detection is to mount next an...

Today's Machine Learning(ML) systems, especially those running in server farms workloads such as Deep Neural Networks, which require billions of parameters and many hours to train a model, consume significant amount energy. To combat this, researchers have been focusing on new emerging neuromorphic computing models. Two models are Hyperdimensional Computing (HDC) Spiking Networks (SNNs), both with their own benefits. HDC has various desirable properties that other Learning (ML) algorithms...

Today's machine learning (ML) systems, running workloads, such as deep neural networks, which require billions of parameters and many hours to train a model, consume significant amount energy. Due the complexity computation topology, even quantized models are hard deploy on edge devices under energy constraints. To combat this, researchers have been focusing new emerging neuromorphic computing models. Two those hyperdimensional (HDC) spiking networks (SNNs), both with their own benefits. HDC...

Multimodal interactions can reduce cognitive load in demanding situations, such as driving [68, 69, 85]. While interfaces are becoming more mainstream vehicles, knowledge of situational and contextual factors which influence user preferences about these is sparse. Through semi-structured interviews with 15 individuals, we take a scenario-specific look at perceptions, expectations, concerns voice, gesture, multimodal within vehicles. We presented participants scenarios where load, social...

Neuromorphic hardware equipped with learning capabilities can adapt to new, real-time data. While models of Spiking Neural Networks (SNNs) now be trained using gradient descent reach an accuracy comparable equivalent conventional neural networks, such often relies on external labels. However, real-world data is unlabeled which make supervised methods inapplicable. To solve this problem, we propose a Hybrid Guided Variational Autoencoder (VAE) encodes event based sensed by Dynamic Vision...

Abstract Achieving personalized intelligence at the edge with real-time learning capabilities holds enormous promise to enhance our daily experiences and assist in decision-making, planning, sensing. Yet, today's technology encounters difficulties efficient reliable edge, due a lack of data, insufficient hardware capabilities, inherent challenges posed by online learning. Over time across multiple developmental phases, brain has evolved efficiently incorporate new knowledge gradually...

Achieving personalized intelligence at the edge with real-time learning capabilities holds enormous promise in enhancing our daily experiences and helping decision making, planning, sensing. However, efficient reliable remains difficult current technology due to lack of data, insufficient hardware capabilities, inherent challenges posed by online learning. Over time across multiple developmental stages, brain has evolved efficiently incorporate new knowledge gradually building on previous...

This live demonstration will show real-time, embedded learning of gestures shown to a dynamics vision sensor in neuromorphic hardware. A multi-layer spiking neural network implemented the Loihi processor partially trained on 11 classes be able learn new by using combination transfer and local synaptic plasticity. Visitors experience real-time they whose data is processed connected chip.

Being very low power, the use of neuromorphic processors in mobile devices to solve machine learning problems is a promising alternative traditional Von Neumann processors. Federated Learning enables entities such as collaboratively learn shared model while keeping their training data local. Additionally, federated secure way because only weights need be between models, private. Here we demonstrate efficacy Neuromorphic benefit from collaborative learning, achieving state art accuracy on...