Maya-Prana: Metabolic Plasticity Budget for Continual Learning in Affective Spiking Neural Networks

摘要

Without Prana, nothing moves, nothing learns, nothing grows. We present Maya-Prana, the ninth and final paper in the Maya Research Series, which completes the Antahkarana by introducing Prana (प्राण) as a metabolic plasticity budget governing how much learning the system can sustain per unit time. Prana is modelled on the Astrocyte-Neuron Lactate Shuttle (ANLS): astrocytes supply lactate fuel to active neurons during sustained synaptic activity; when metabolic demand exceeds supply, plasticity degrades. In Maya-Prana, Prana is a scalar budget that depletes under gradient load proportional to gradient magnitude and neural activity, recovers during low-activity batches modulated by Vairagya, partially restores at task boundaries (sleep analogue), and gates the effective learning rate per batch: effective_lr = base_lr × prana × (0.5 + buddhi × 0.5). The biological calibration (PRANA_COST_RATE=0.002315, the ORCID magic number) produces a system in which Prana maintains full budget (1.0000) throughout all 10 tasks — consistent with the ANLS literature, which confirms that the astrocytic metabolic supply does not fail under standard cognitive load. A six-condition ablation on Split-CIFAR-100 (10 tasks, seed=42) establishes four findings. Condition C (fixed Prana=1.0) and Condition D (canonical) are near-equivalent — AA=12.02% versus 12.72% — confirming Prana resilience as the primary result. Condition B (Prana without full Antahkarana) produces the worst result (AA=10.33%, Pruned=91.85%), proving that Prana cannot substitute for the integrated Antahkarana. Condition E (PRANA_COST_RATE=0.008, 3.5× canonical) still never depletes Prana, confirming ANLS robustness under accelerated metabolic demand. Condition F (no Buddhi modulation, fixed EffLR=0.0075) produces the unexpected best result: AA=13.68%, BWT=−51.20%, Pruned=46.93%, revealing that the Buddhi warm-up schedule penalises early task consolidation — an honest finding about the Buddhi-Prana interaction term, reported as discovered. We confirm, for the ninth consecutive paper, that Bhaya quiescence under replay is a series-level constant — the Bhaya Quiescence Law. Buddhi's S-curve consolidation gate is confirmed as architecturally deterministic across all six conditions. An embodied demonstration of the full Antahkarana running on a PiCar-X robotic platform with Raspberry Pi 5 is available on YouTube (see Section 5.9). Code, ablation scripts, interactive bilingual dashboard, and steganographically signed figures are available at github.com/venky2099/Maya-Prana. Dashboard: https://venky2099.github.io/Maya-Prana/docs/maya_prana_dashboard.html FAQ: https://venky2099.github.io/Maya-Prana/docs/faq.html Across nine papers, we have demonstrated the computational maturation of a mind. Website: https://venky2099.github.io/ Series: Part of the Maya Research Series — 12 papers implementing the Advaita Vedantic Antahkarana as computational primitives in spiking neural networks. Bhaya Quiescence Law and Buddhi S-Curve Determinism confirmed across all papers. Links: GitHub Repository (private — to request access: email research@nexuslearninglabs.in with subject Code Access Request — Maya-Prana and your research context) | Interactive Dashboard | FAQ | Full Series Index — venky2099.github.io Nexus Learning Labs, Bengaluru · UDYAM-KR-02-0122422 · BHASKAR IN-0526-9452JSORCID: 0000-0002-3315-7907 · VAIRAGYA_DECAY_RATE = 0.002315 (embedded in all canonical hyperparameters)Canary: MayaNexusVS2026NLL_Bengaluru_Narasimha