Our research

Empowering discovery. Respecting human judgement.

• [May 2026 – present] – We work on General Pretrained Kernels (GPK) as a semantic substrate layer for Lifelong Embodied AI.
• [May 2026, paper, code] – We provide a unifying view on neural computation, bridging ideas like gradient smoothing, B-cos-style alignment, feature accentuation, robust optimization and normalization effects. We introduce Semantic Pullbacks as an effective explanation method for deep models.
• [March 2024, paper, code] – We introduce Semantic Features as a candidate building block for transparent neural networks that are both self-explainable and adversarially robust.