△ Mechanism confirmed, baseline not beaten 2026

Gram-Whitened Directional Pooling

Implementation & benchmark of arXiv:2608.12794 — Point Feature Descriptor via Directional Partition of Unity on Maps

Usefulness7/10

Difficulty5/10

Novelty6/10

Architecture Accuracy Stability Generalization Embedding Cnn Transformer Loss Functional analysis Linear algebra Geometry Approximation theory

Source paper: Point Feature Descriptor via Directional Partition of Unity on Maps arXiv:2608.12794 ⓘ · analyzed Sep 1, 2026

AI-generated research hypothesis, automatically tested. Not peer-reviewed.

Idea description

Represent local feature channels as a smooth directional signal and aggregate them with a partition-of-unity family of learnable spherical atoms instead of hard angular bins. Use the atom Gram matrix to whiten the descriptor and add a projected-energy loss, so the network is rewarded for retaining information in the directional subspace rather than merely producing large correlated channel responses.

Formulas

$$\phi_i(f)=\langle f,w_i\rangle_{L^2(\mathbb S^{d-1})}=\int_{\mathbb S^{d-1}}f(\theta)w_i(\theta)\,d\sigma(\theta),\qquad \sum_{i=1}^{n}w_i(\theta)=1,\quad w_i(\theta)\geq 0.$$

$$w_i(\theta)=\frac{\exp(d_i^\top\theta/\varepsilon)}{\sum_{j=1}^{n}\exp(d_j^\top\theta/\varepsilon)},\qquad d_i\in\mathbb S^{d-1},\ \varepsilon>0.$$

$$G_{ij}=\langle w_i,w_j\rangle_{L^2}=\int_{\mathbb S^{d-1}}w_i(\theta)w_j(\theta)\,d\sigma(\theta),\qquad P_nf(\theta)=\sum_{i,j=1}^{n}w_i(\theta)(G^\dagger)_{ij}\phi_j(f).$$

$$|f|_{\mathcal D,n}^{2}=\|P_nf\|_{L^2}^{2}=\phi(f)^\top G^\dagger\phi(f),\qquad |f|_{\mathcal D,n}\longrightarrow\|f\|_{L^2(\mathbb S^{d-1})}\ \text{as }n\to\infty.$$

Mathematical statement

The descriptor has coordinates $\phi_i(f)=\langle f,w_i\rangle_{L^2(\mathbb S^{d-1})}=\int_{\mathbb S^{d-1}}f(\theta)w_i(\theta)\,d\sigma(\theta)$, where $f$ is a directional feature signal, $w_i$ is the $i$-th nonnegative weight, and $\sigma$ is surface measure. The weights satisfy the partition of unity $\sum_{i=1}^n w_i(\theta)=1$, $w_i(\theta)\ge 0$. For unit directions $d_i\in\mathbb S^{d-1}$ and temperature $\varepsilon>0$, use $w_i(\theta)=\exp(d_i^\top\theta/\varepsilon)/\sum_{j=1}^n\exp(d_j^\top\theta/\varepsilon)$. Define the Gram matrix $G\in\mathbb R^{n\times n}$ by $G_{ij}=\langle w_i,w_j\rangle_{L^2}=\int w_i(\theta)w_j(\theta)\,d\sigma(\theta)$, and descriptor vector $c_i=\phi_i(f)$. The orthogonal projection onto $W_n=\operatorname{span}\{w_i\}$ is $P_nf(\theta)=\sum_{i,j=1}^n w_i(\theta)(G^\dagger)_{ij}c_j$, where $G^\dagger$ is the Moore-Penrose pseudoinverse. Its induced seminorm is $|f|_{\mathcal D,n}^2=\|P_nf\|_{L^2}^2=c^\top G^\dagger c$, and this converges to $\|f\|_{L^2(\mathbb S^{d-1})}^2$ as the directional family becomes total. In a network, this gives a principled decorrelated descriptor and a loss measuring retained directional energy.

Implementation notes

Integrate this into a CNN or vision-transformer block immediately after a local feature tensor has been converted into directional samples. For each spatial token x, construct K unit directions theta_k from fixed image-plane offsets, 3D point-neighborhood offsets, or a learned normalized direction head. Let F_{x,k} in R^C be the feature at direction theta_k. Maintain n learnable unit atoms d_i, and compute W_{k,i}=softmax_i(d_i^T theta_k/epsilon). Aggregate descriptor channels as c_{x,i}=sum_k q_k F_{x,k} W_{k,i}, where q_k are quadrature weights summing to one. Precompute or update the Gram matrix G_{ij}=sum_k q_k W_{k,i}W_{k,j}. Stabilize it with G_lambda=G+lambda I, solve G_lambda z_x=c_x by Cholesky, and feed the whitened descriptor z_x to the next layer; stop gradients through G initially, then optionally update it with an exponential moving average. Add L_energy=||c_x||_2^2-c_x^T G_lambda^{-1}c_x only if the target task benefits from preserving directional energy, or use c_x^T G_lambda^{-1}c_x as an auxiliary representation norm. The paper supplies the projection and Gram formulas; the engineer estimates integrals by the finite directional quadrature above and chooses lambda empirically. First test on CIFAR-100 or ModelNet40 using a small ResNet with four angular atoms versus hard directional binning and ordinary softmax pooling. Measure validation accuracy, sensitivity to rotated inputs, atom-correlation condition number, and training-loss variance. Success means better rotation robustness or accuracy at equal FLOPs, with a lower Gram condition number and no instability when epsilon is annealed.

Verification

Mechanism confirmed, baseline not beaten

Stage 1 · Toy mechanism gate: Passed ✓

Stage 2 · Mechanism transferred to benchmark: Confirmed ✓

Stage 2 · Practical benchmark result: No measurable effect

Methodology: 8 paired seeds, permutation test (p<0.05) against a learning-rate-tuned baseline. How verification works

Stage 1 — Mechanism check agent confidence 8/10

Built a NumPy MVP for softmax directional atoms, finite-quadrature Gram matrices, Gram whitening, projected energy, and ridge stabilization. The checks confirmed partition-of-unity error 3.3e-16, projection energy 0.2967 <= signal energy 0.845, projection reconstruction identity error 9.4e-16, monotone energy reduction across lambda, and asymptotic lambda*E=0.0342766 versus predicted c^T c=0.0342827. The dense atom sweep approached full signal energy (0.8450), validating the totality-style trend, but the toy classification accuracy was identical for raw pooling and whitening (0.5854), so no downstream performance win was observed.

Agent confidence: 8/10

Baseline: Raw soft pooling nearest-centroid accuracy: 0.5854; signal energy: 0.8450; Gram condition number: 37.87.
Idea: Gram-whitened nearest-centroid accuracy: 0.5854; projection energy: 0.2967; projection energy fraction: 0.3512; lambda=1000 asymptotic lambda*E: 0.0342766 vs predicted 0.0342827.

Limitations: Only a small synthetic 2D directional signal experiment was tested; no CNN, CIFAR-100/ModelNet40, learned atoms, rotation-robustness benchmark, GPU execution, FLOP measurement, or training-loss variance study was performed. The atom-count energy trend was sampled and approached full energy but was not strictly monotone at every intermediate count.

How to run: python3 gram_directional_pooling.py

Stage 2 — Benchmark vs. tuned baseline (arXiv:2608.12794) Sep 2, 2026

No significant difference vs. the tuned baseline (+0.4%, p=0.8129, wins 4 / 8 paired seeds).

Benchmark: Vision (image classification)
Model: cnn_small
Paired seeds: 8
Baseline mean: 0.795
Idea mean: 0.7981
Effect (Δ): 0.0031 (+0.4%; negative = idea better)
Wins: 4 / 8 paired seeds
p-value: 0.8129 (permutation test, 20 000 shuffles)
Smallest detectable effect: ±3.0%
Mechanism: Confirmed ✓
Practical effect: No effect
Baseline tuning: swept over 3 configs

Limitations:

Tested only the built-in CIFAR-10 subset vision track, cnn_small architecture, 8 epochs, 400 training and 200 test samples per seed, and three learning rates with fixed epsilon/lambda choices. Rotation robustness, larger datasets, alternative atom counts, annealed epsilon, and FLOP/throughput measurements were not tested.

How to run: python3 stage2_bench.py

Verdict computed by deterministic test code from paired-seed statistics — not by the language model.

Stage 2 — Benchmark Sep 1, 2026 latest Failed ✗

Agent confidence: 9/10

Implemented Gram-whitened directional pooling as an end-to-end CNN readout with learnable normalized atoms, soft partition-of-unity weights, finite Gram estimation, and ridge whitening. On the required 8-seed vision benchmark, the idea was slightly worse than the tuned baseline (error 0.798125 vs 0.795000; paired delta +0.003125; permutation p=0.8129), so there is no significant win. The trained-model mechanism signature was confirmed: partition-of-unity error was 1.19e-7 and projected energy 0.07936 was below pooled feature energy 0.08952, but this did not translate into task improvement.

Baseline: vision/cnn_small error mean=0.79499997, std=0.02091650, best lr=0.001, epochs=8.
Idea: Gram-whitened directional CNN error mean=0.79812498, std=0.03741134, best tested lr=0.003, epsilon=0.35, lambda=0.01, epochs=8; paired delta=+0.00312500, p=0.8129, 4/8 seed wins.

Benchmark result