Use the future to build the present
Neuromorphic Benchmarking
Comment
Stakeholder Type
,
1Quantum Revolution& Advanced AI2HumanAugmentation3Eco-Regeneration& Geo-Engineering4Science& Diplomacy1.11.21.31.42.12.22.32.43.13.23.33.43.54.14.24.34.44.5HIGHEST ANTICIPATIONPOTENTIALAdvancedArtificial IntelligenceQuantumTechnologiesBrain-inspiredComputingBiologicalComputingCognitiveEnhancementHuman Applications of Genetic EngineeringRadical HealthExtensionConsciousnessAugmentation DecarbonisationWorldSimulationFuture FoodSystemsSpaceResourcesOceanStewardshipComplex Systems forSocial EnhancementScience-basedDiplomacyInnovationsin EducationSustainableEconomicsCollaborativeScience Diplomacy
1Quantum Revolution& Advanced AI2HumanAugmentation3Eco-Regeneration& Geo-Engineering4Science& Diplomacy1.11.21.31.42.12.22.32.43.13.23.33.43.54.14.24.34.44.5HIGHEST ANTICIPATIONPOTENTIALAdvancedArtificial IntelligenceQuantumTechnologiesBrain-inspiredComputingBiologicalComputingCognitiveEnhancementHuman Applications of Genetic EngineeringRadical HealthExtensionConsciousnessAugmentation DecarbonisationWorldSimulationFuture FoodSystemsSpaceResourcesOceanStewardshipComplex Systems forSocial EnhancementScience-basedDiplomacyInnovationsin EducationSustainableEconomicsCollaborativeScience Diplomacy

Sub-Field:

1.3.4Neuromorphic Benchmarking

Neuromorphic chips should be well-suited to situations where information and demands are fluid, energy consumption has to be low and adaptation to novel situations is required. But as yet there is no “killer app” for early brain-like computing that might demonstrate its potential, and no agreed universal standard for benchmarking how well the field, or an individual device, is progressing. This is going to be a crucial part of the research effort, since it will provide conceptual understanding, incentive for progress and rewards for investment and innovation.14
It is important for the field to begin demonstrations of its potential for fast, low power processing.15 However, it is also important not to try to compete with deep learning algorithms of artificial intelligence research programmes, which have received significant input on extremely specific capabilities, such as machine vision. There are many other sensory modalities, such as hearing and touch, that will be technologically important, and just as important is the ability to do fast, real-time, self-contained sensory processing, rather than rely on connections with a cloud-based data centre.

Future Horizons:

×××

5-yearhorizon

Standardised benchmarking tools emerge

Researchers agree a set of simulators or standard robotic platforms for benchmarking progress and accelerating promising candidate architectures. Aware of the pitfalls encountered in machine vision and deep learning, they resist the pressure to optimise their systems for achieving benchmarks over useful real-world tasks.

10-yearhorizon

Real-world testing accelerates progress

Hand-held devices that embody insect-level intelligence and learning become ubiquitous, and real-world benchmarking brings commercial pressures that accelerate progress. Energy-efficient, low-latency neuromorphic systems can be tested against, and begin to outperform, data centre-connected deep learning algorithms for tasks such as speech recognition.

25-yearhorizon

Human-machine interfaces allow subjective testing and user review

Bio-compatible neuromorphic machine interfaces, integrated with the nervous system, become widely available. This creates a product marketplace based on user experience, further accelerating progress.

Neuromorphic Benchmarking - Anticipation Scores

How the experts see this field in terms of the expected time to maturity, transformational effect across science and industries, current state of awareness among stakeholders and its possible impact on people, society and the planet. See methodology for more information.

GESDA Best Reads and Key Resources