The Mechanics of Forgetting E1 | Neural Pruning — Biological Feasibility of Structural Distillation
Description
【要旨】
発達期および学習後の神経系では、シナプス結合の30-50%が選択的に除去される(神経刈り込み)。この現象は従来、効率化・ノイズ除去・競合選択として説明されてきたが、なぜ削減後に構造が明瞭化し、性能が向上するのかは未解明であった。本稿は、神経刈り込みを構造的蒸留(𝓕)の枠組みで再解釈する。C0で定義されたBP=(A,C,φ;K)——構造軸A、主張ノードC、判定関数φ、制約集合K——を神経系に写像し、刈り込み前後の圧縮率(CR)と構造保存度(SP)を測定した。in-silico実験(MNIST分類、n=3)において、刈り込み率α=0.4-0.6の範囲でCR↑とSP≥0.60の両立への接近が観察された(α=0.6: CR=0.60, SP=0.64)。この傾向は価値中立条件(V=neutral)でも維持され、価値感度VSは低域(0.18)に留まった。過剰刈り込み(α>0.7)ではSPが急落し(0.43)、一般化能力GENも負の値を示した。生物学的データ(Allen V1)の再解析でも、発達期の刈り込み(35%減少)において抽象度の上昇と性能向上が確認された。これらの結果は、刈り込みが単なる削減ではなく、制約Kに導かれた構造再編——すなわち構造的蒸留——として機能しうることを示唆する。本稿は概念実証段階であり、厳密な測定系の確立はC3以降に委ねる。
キーワード: 神経刈り込み、構造的蒸留、シナプス可塑性、圧縮率、構造保存度、価値中立性
【Abstract】
During development and post-learning periods, neural systems selectively eliminate 30-50% of synaptic connections (synaptic pruning). While this phenomenon has been traditionally explained through efficiency optimization, noise reduction, and competitive selection, it remains unclear why structural clarity and performance improve after pruning. This study reinterprets neural pruning through the framework of structural distillation (𝓕). We mapped the C0-defined BP=(A,C,φ;K)—comprising structural axes A, claim nodes C, coherence function φ, and constraint set K—onto neural systems, measuring compression rate (CR) and structure preservation (SP) before and after pruning. In in-silico experiments (MNIST classification, n=3), we observed an approach to dual achievement of CR↑ and SP≥0.60 within pruning rate α=0.4-0.6 (α=0.6: CR=0.60, SP=0.64). This trend persisted under value-neutral conditions (V=neutral), with value sensitivity VS remaining low (0.18). Over-pruning (α>0.7) resulted in SP collapse (0.43) and negative generalization capacity (GEN<0). Reanalysis of biological data (Allen V1) confirmed increased abstractness and performance improvement during developmental pruning (35% reduction). These results suggest that pruning functions not merely as reduction, but as constraint-guided structural reorganization—i.e., structural distillation. This study represents conceptual validation; rigorous measurement protocols will be established in C3 onwards.
Keywords: synaptic pruning, structural distillation, synaptic plasticity, compression rate, structure preservation, value neutrality
Files
2025-124-忘却の構造力学 E1.pdf
Additional details
Additional titles
- Translated title
- 忘却の構造力学 E1|神経刈り込み —— 構造的蒸留の生物学的実装可能性
Related works
- Cites
- 10.5281/zenodo.17416598 (DOI)
- Is supplement to
- 10.5281/zenodo.17416598 (DOI)
References
- [1] Huttenlocher PR, Dabholkar AS. (1997) Regional differences in synaptogenesis in human cerebral cortex. J Comp Neurol, 387(2):167-178.
- [2] Holtmaat A, Svoboda K. (2009) Experience-dependent structural synaptic plasticity in the mammalian brain. Nat Rev Neurosci, 10(9):647-658.
- [3] Casey BJ, Tottenham N, Liston C, Durston S. (2005) Imaging the developing brain: what have we learned about cognitive development? Trends Cogn Sci, 9(3):104-110.
- [4] Karbowski J. (2014) Constancy and trade-offs in the neuroanatomical and metabolic design of the cerebral cortex. Front Neural Circuits, 8:9.
- [5] Turrigiano GG. (2008) The self-tuning neuron: synaptic scaling of excitatory synapses. Cell, 135(3):422-435.
- [6] Hua JY, Smith SJ. (2004) Neural activity and the dynamics of central nervous system development. Nat Neurosci, 7(4):327-332.
- [7] Bassett DS, Sporns O. (2017) Network neuroscience. Nat Neurosci, 20(3):353-364.
- [8] Petanjek Z, et al. (2011) Extraordinary neoteny of synaptic spines in the human prefrontal cortex. PNAS, 108(32):13281–13286.
- [9] Gogtay N, et al. (2004) Dynamic mapping of human cortical development during childhood through early adulthood. PNAS, 101(21):8174–8179.
- [10] Bourgeois JP, Rakic P. (1993) Changes of synaptic density in the primary visual cortex of the macaque monkey from fetal to adult stage. J Neurosci, 13(7):2801–2820.
- [11] Rakic P, Bourgeois JP, Goldman-Rakic PS. (1994) Synaptic development of the cerebral cortex: implications for learning, memory, and mental illness. Prog Brain Res, 102:227–243.
- [12] Zhang LI, et al. (2005) Topography and synaptic shaping of direction selectivity in primary auditory cortex. Nature, 424(6945):201–205.
- [13] Yang G, et al. (2009) Stably maintained dendritic spines are associated with lifelong memories. Nature, 462(7275):920–924.
- [14] Luna B, et al. (2015) An integrative model of the maturation of cognitive control. Annu Rev Neurosci, 38:151–170.
- [15] Chaudhuri R, et al. (2015) A large-scale circuit mechanism for hierarchical dynamical processing in the primate cortex. Neuron, 88(2):419-431.
- [16] Ringach DL. (2004) Mapping receptive fields in primary visual cortex. J Physiol, 558(3):717-728.
- [17] Tononi G, Cirelli C. (2014) Sleep and the price of plasticity: from synaptic and cellular homeostasis to memory consolidation and integration. Neuron, 81(1):12-34.
- [18] de Vries SEJ, et al. (2020) A large-scale standardized physiological survey reveals functional organization of the mouse visual cortex. Nat Neurosci, 23(1):138-151.
- [19] Fair DA, et al. (2009) Functional brain networks develop from a "local to distributed" organization. PLoS Comput Biol, 5(5):e1000381.
- [20] Tishby N, Zaslavsky N. (2015) Deep learning and the information bottleneck principle. IEEE Information Theory Workshop, 1–5.
- Hideki. (2025). 忘却の構造力学 C0|原理 —— 構造的蒸留と価値駆動再生成 (プレプリント 2025-122). Zenodo. https://doi.org/10.5281/zenodo.17416598