Appears in a recent preprint, that Basharat Ali has been running Akida, with others, as part of his / her work on cybersecurity.
Akida gives some pretty good results.
Kinda craps on NVIDiA.
Neuromorphic Quantum Adversarial Learning
(NQAL): A Bio-Inspired Paradigm for DNS over HTTPS Threat Detection
Basharat Ali
Nanjing University
Research Article
Keywords: Network Security, NQAL in Network Security, Network Protocols, Enhancing Network Security,
Enhancing DoH Protocol Security, Threats Detection in Encrypted Network, Cyber Attacks Detections
Posted Date: April 30th, 2025
Abstract Excerpt:
To overcome these complex issues, this work proposes a new architecture—Neuromorphic Quantum Adversarial Learning (NQAL)—a bio-inspired, zero-knowledge-supported detection
mechanism combining spiking neural networks (SNNs), quantum noise injection (QNI), and federated swarm intelligence to immunize, rather than detect, DoH-based attacks.
The method relies on a neuromorphic model employing Dynamic Spiking Graph Attention (DSGAT) and Spike-Timing-Dependent Plasticity (STDP) to encode encrypted traffic as dynamic spike trains to enable ultra-fast, energy-efficient inference on processors such as Intel Loihi and BrainChip Akida
Experiment set up Except:
Experiments were carried out on neuromorphic hardware platforms such as Intel Loihi 2 and
BrainChip Akida that provide sub-millisecond latency with low-power event-driven processing characteristics.
Akida results related Excerpt:
Table 5: Hardware Deployment Metrics
Platform Accuracy Latency Power Throughput
GPU (NVIDIA V100) 89.2% 3.1 ms 45 W 1,200 QPS
TPUv4 91.5% 2.8 ms 32 W 1,500 QPS
Loihi 2 98.7% 0.9 ms 4 W 9,800 QPS
Akida 99.1% 0.7 ms 3 W 12,400 QPS
Outcome of Table 5:
Hardware Installation Metrics presents the excellent performance of our neuromorphic hardware solutions towards accomplishing peak performance for DoH security systems.
When comparing Loihi 2 and Akaida to GPU platforms and TPU platforms depicts easily how changing towards neuromorphic chips invokes important boosts in terms of both accuracy and efficiency. Both the GPU (NVIDIA V100) and TPUv4 initiated with low performance at 89.2% and 91.5% accuracy, respectively, but
when executed on Loihi 2, accuracy jumped dramatically to 98.7%, and a further improved 99.1% on Akida.
This increase in accuracy is accompanied by a drastic reduction in latency, from 3.1 ms for GPU to 0.7 ms for Akida, illustrating the real-time processing capability of the neuromorphic hardware.
Besides this, the power usage of the
Loihi 2 and Akida platforms—4 W and 3 W respectively—is a brilliant power efficiency against traditional GPU-based systems consuming 45 W. Throughput is also dramatically increased, with Akida being able to support 12,400 QPS, in strong contrast to the GPU’s 1,200 QPS.
Such results justify the single value of neuromorphic hardware as an approach for energy-efficient high-performance DoH anomaly detection and prove how
our new approach beats current systems and becomes the future standard for real-time system encrypted traffic protection[14].
Full paper
HERE
Top 20 Most Popular State Space / Sequential Models (Including LLMs & Modern Hybrids)
Emerging or Specialized Models (Ranks 11–20)
Summary
