Whitepapers & onderzoek
Technische inzichten in binaire neurale netwerken en efficiënte AI-architecturen
Hybrid Binary Intelligence: Neural Networks, Constraints, and Multi-Paradigm Architectures
Binary neural networks replace floating-point arithmetic with single-bit operations, enabling neural network inference at a fraction of conventional computational cost. However, pure binary networks struggle to match the accuracy of full-precision models. This paper presents a hybrid architecture that combines binary neural networks with constraint-based reasoning, hyperdimensional computing, and adaptive precision to achieve greater than 99% accuracy on domain-specific tasks.
Door Marc Filipan
Permuted Agreement Popcount: Structural Similarity in Binary Vector Spaces
Hamming distance provides efficient binary vector comparison but fails to capture structural coherence in matching patterns. Two binary vectors with identical Hamming similarity can exhibit fundamentally different spatial arrangements: one with matches forming coherent blocks, another with matches scattered randomly. This structural blindness creates problems in applications where spatial arrangement matters (computer vision, document similarity, binary neural networks).
Door Marc Filipan
AI Safety: Beyond the Theater - Real Solutions to Current Threats
Most AI companies claim GDPR compliance, robust security, and EU AI Act readiness. Yet prompt injection attacks succeed against over 90% of commercial LLM platforms. PII leaks from systems displaying compliance badges. Model extraction proceeds undetected. The gap between AI safety claims and actual implementation has become a credibility crisis.
Door Marc Filipan
Meld je aan voor de Dweve-wachtlijst
Krijg vroege toegang tot AI die je privacy, de planeet en je budget respecteert.
Voordelen van vroege toegang
Snel proces
Duurt minder dan 2 minuten • Geen creditcard nodig
Blijf op de hoogte van ons onderzoek
Ontvang een melding wanneer we nieuwe whitepapers en onderzoeksbevindingen publiceren