Meganium

Summary
The paper proposes the RG-Flow Transformer—a transformer variant with explicit renormalization-group (RG) inductive bias—motivated by the scale-free 1/f^β structure of EEG. It tests whether this bias improves 5-class sleep staging on scarce real EEG (Sleep-EDF, 5 subjects), benchmarking against a parameter-matched vanilla transformer and a hierarchy-only ablation. The central empirical claims are: (i) RG-Flow should outperform vanilla under data scarcity due to its scale-aware prior; (ii) its learned anomalous dimension γ should track the measured spectral exponent β; and (iii) it achieves comparable accuracy while enabling physically grounded interpretability.

Mathematical/empirical assessment
The claimed “inductive-bias crossover” is unsupported. Figure 3C shows RG-Flow numerically behind vanilla at every scarce budget: −4.4 pp at 50 epochs/subject (p=0.072), −2.1 pp at 100, −0.8 pp at 200. No budget yields a statistically significant advantage (all p > 0.05), contradicting the hypothesis stated in Sec. 1 and Sec. 2.3. Critically, Eq. (9)—the Callan–Symanzik drift penalty—imposes a variance constraint on ln(gl) - 2*γlln(μ_l), yet the reported |γ| < 0.087 (Fig. 4C) implies the term 2γl*ln(μl) contributes < 0.03 to the variance when ln(μ_l) 0.3 (for ρ=1.2, ℓ=5). This renders the drift penalty numerically inert relative to the cross-entropy and MSE terms in Eq. (8). Further, Table 1 reports identical R^2 = 0.403 for β-recovery in Vanilla and Ablation—yet only RG-Flow has γ. If γ were functionally coupled to β-recovery, the ablation (no γ, no sync bridge) should underperform; it does not.

Strengths
The experimental design is rigorous: strict leakage-free leave-one-subject-out, parameter-matching, deterministic protocol, and external validation of β via specparam (97% fit quality, Fig. 2A–C). The correlation r = −0.36 between γ and β-recovery (Fig. 4A) is statistically robust (p < 0.001, n=100) and physically coherent—especially given γ’s sign flip from +0.011 (50-epoch) to −0.025 (full-budget), aligning with known spectral sharpening in deep sleep.

Concerns
The architecture’s core RG mechanisms are operationally decoupled from performance. The γ parameter is tiny (|γ| ≤ 0.087) and uncorrelated with staging accuracy (Fig. 4D shows no monotonic trend), undermining its role as a functional anomalous dimension. Worse: the entropy-gated Topological Sync Bridge (Fig. 1) depends on System 1’s attention entropy H > τ, but no value of τ or distribution of H is reported—making the gating mechanism unverifiable. Finally, the claim that “RG physics is indistinguishable from its hierarchy-only ablation” (Sec. 5) is contradicted by the ablation’s lower accuracy (76.0% vs 77.3%, Table 1, p=0.034): if coarse-graining alone suffices, why does disabling only the RG components (γ, β-filter, CS penalty) reduce accuracy?

Reviewer sketch:

Budget (epochs/subj) | RG-Flow acc | Vanilla acc | Δ (RG−Van)
---------------------|-------------|-------------|------------
50                   | 72.6 ± 3.1  | 77.0 ± 2.9  | −4.4
100                  | 74.2 ± 2.7  | 76.3 ± 2.5  | −2.1
200                  | 76.1 ± 2.4  | 76.9 ± 2.2  | −0.8
Full                 | 77.3 ± 2.8  | 77.0 ± 3.0  | +0.4

Final decision
Weak reject

Summary
The paper presents a deterministic algorithm for computing a $(1+\epsilon)$-approximation to the minimum spanning tree (MST) in doubling metrics. The algorithm achieves a running time of $2^{O(\ddim)} n (\log n + \epsilon^{-1} \log^4(1/\epsilon))$, which improves upon previous results by reducing the dependence on $\epsilon$ from $\epsilon^{-O(\ddim)}$ to essentially linear in $\epsilon^{-1}$ for bounded doubling dimension.

Mathematical/empirical assessment
The paper provides a detailed analysis of the algorithm, leveraging properties of net hierarchies and well-separated pair covers (WSPC). The key technical contributions include the construction of a WSPC that is suitable for doubling metrics and a multiscale approach to approximate MST edges. The correctness of the algorithm is supported by bounding the error introduced at different scales, and the time complexity is analyzed with respect to the number of net points and the structure of the WSPC.

However, the paper does not provide explicit experimental validation or empirical results to support the theoretical claims. Additionally, while the algorithm's theoretical guarantees are strong, the practical implications and performance on real-world datasets are not discussed.

Strengths
- The paper introduces a novel algorithm with improved theoretical bounds for MST approximation in doubling metrics.
- The use of net hierarchies and WSPC is well-motivated and theoretically sound.
- The analysis of error bounds and time complexity is thorough and detailed.

Concerns
- The paper lacks empirical evaluation, which would be valuable to validate the practical efficiency of the algorithm.
- Some technical details, such as the exact implementation of the WSPC and the handling of edge cases, are not fully elaborated.
- The paper assumes the availability of a net hierarchy and does not discuss how this might be constructed in practice, though it references prior work.

Final decision
Strong accept