Independent AI Systems Researcher — boundary testing, recursive learning systems, and alternative architectures.
I build CPU-runnable experimental AI systems that treat failure as signal, not noise. My research explores where learning systems break under constrained compute, weak evaluators, distribution shift, brittle abstractions, benchmark leakage, and long-range dependency pressure — then turns those breakdowns into better mechanisms for search, validation, memory, and generalization.
- DeepNeural-AutoExploration — recursive adaptation loops with non-leaking episodic memory, operator-program synthesis, validation-only evaluation, evaluator evolution, and failure-to-rule compression.
- RSI-NAS-Attention-Free — neural architecture search for attention-free sequence models, including routing, spectral propagation, dynamic gating, hierarchical pooling, and field-based alternatives to quadratic attention.
- OMEGA-THDSE — topological, hyperdimensional, and symbolic system experiments for structural representation and reasoning.
- attention-free-sequence-model — compact experiments around non-attention sequence computation.
- Failure-driven mechanism discovery
- Boundary-condition testing and anti-cheat evaluation
- Recursive and self-evolving learning systems
- Attention-free long-range sequence modeling
- Structural memory and operator-program mutation
- CPU-constrained experimental AI systems
Email: sunghunkwag@gmail.com