Generative Latent-Space Bridging for Adaptive HCI
Developing a gradient-free translator for multi-finger gesture typing. Couples frozen Touch and Text VAEs, bridging their latent spaces through vector field dynamics. Target words act as semantic attractors, enabling continuous trajectory decoding and one-shot personalization through magnetic latent interpolation.
Elastic Homeostasis in Spiking Neural Networks
Can learning emerge from adaptive equilibrium points rather than synaptic weights? Testing a novel hypothesis where neurons maintain stable firing patterns but can shift to new equilibria under sustained perturbation. Implemented 100k neuron simulation using homeostatic Izhikevich model.
Minimal World Model Formation
Can an AI develop functional world understanding from a single temporal signal? Challenging the assumption that models need pre-structured data. Testing if world models can bootstrap from one value changing over time through predictive coding and continuous architectural plasticity.