Pith. sign in

REVIEW 5 major objections 5 minor 1 cited by

The Geometry of ReLU Networks through the ReLU Transition Graph

T0 review · 5 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The capacity of a ReLU network is bounded by the diameter of its ReLU Transition Graph.

desk verdict The central VC-dimension bound is false: a two-neuron ReLU network in R^2 has RTG diameter 2 but VC dimension at least 3, so the paper's main theorem collapses immediately. read the letter →

arxiv 2505.11692 v2 pith:6D5LXQU7 submitted 2025-05-16 cs.LG cs.AIcs.NE

Sahil Rajesh Dhayalkar This is my paper · ORCID
classification cs.LGcs.AIcs.NE MSC 68T0752C35
keywords ReLUnetworkslinearregionsactivationpatternsTransitionGraphdiameterVCdimensionnetworkcompressionpiecewise-linearfunctions
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper introduces a graph called the ReLU Transition Graph (RTG), whose nodes are the distinct linear regions of a ReLU network and whose edges connect regions that differ by flipping one neuron. The paper argues that this graph is connected, that its diameter bounds the network's VC dimension, and that its degree and entropy statistics govern generalization and compressibility. If these claims hold, they would give a purely structural, parameter-count-free account of network capacity and a principled basis for pruning. The paper backs the theorems with experiments on two-dimensional inputs that it reads as confirming connectivity, diameter bounds, degree sparsity, and compression error.

What carries the argument

The load-bearing object is the ReLU Transition Graph (RTG): an undirected graph whose vertices are the network's linear regions, each identified with a binary activation pattern, and whose edges join regions sharing a $(d-1)$-dimensional boundary, which the paper equates with a Hamming distance of exactly one between activation patterns, meaning a single neuron flips. All the main theorems are statements about this graph: connectivity is a property of the region complex, the VC bound is a diameter statement, the entropy bound is a function of average degree, and the compression theorem is a statement about low-degree nodes. The graph converts questions about continuous piecewise-linear functions into combinatorial questions about node sets, walks, and degrees.

What would settle it

Densely sample the activation patterns of a two-hidden-layer ReLU network on the plane and, for every pair of patterns at Hamming distance 1, check whether their corresponding regions actually share a face; a single counterexample pair would invalidate the edge rule, the connectivity theorem, and the diameter bound built on it.

Watch

Extended reading notes

Core claim

The central claim is Theorem 3: for a fully connected ReLU network, the VC dimension of its function class is bounded above by the diameter of its ReLU Transition Graph, $\mathrm{VC}(\mathcal{F}) \le \mathrm{diam}(G)$. The diameter is the longest shortest path between any two nodes, i.e., the maximum number of single-neuron flips required to move from one linear region to another. The paper also claims the RTG is always connected under its assumptions (Theorem 2), that the number of regions grows at most like $\sum_{i=0}^{d} \binom{n}{i}^{L}$ (Theorem 1), and that pruning low-degree nodes yields a surrogate function $f_{\mathrm{core}}$ whose uniform deviation from $f$ is arbitrarily small outside a measure-zero boundary set (Theorem 4). Taken together, these are presented as a unified graph-theoretic treatment of expressivity, generalization, and compression.

Load-bearing premise

The paper assumes that every neuron's activation boundary is a single flat hyperplane through the whole input space, even though for all but the first layer these boundaries are bent, piecewise-linear surfaces.

Editorial extensions

If this is right

  • If $\mathrm{VC}(\mathcal{F}) \le \mathrm{diam}(G)$ holds, capacity can be controlled by the longest single-flip path in the region graph, without counting parameters or measuring weight norms.
  • Connectivity of the RTG means that every linear region is reachable from every other by a finite sequence of single-neuron flips, so activation space has no isolated pockets.
  • The entropy lower bound $\mathcal{H}(G) \ge \log(d_{\mathrm{avg}}+1)$ connects uniform region volumes to high average degree, offering a graph-theoretic proxy for generalization that could be computed from data.
  • The degree-sparsity lemma and the compression theorem together imply that a large fraction of low-degree, low-volume regions can be deleted with bounded uniform error, giving a structural explanation for why pruning often succeeds.
  • The size bound $\sum_{i=0}^{d} \binom{n}{i}^{L}$ reproduces the known scaling of region counts—polynomial in width and exponential in depth—within the graph framework.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • One immediate testable extension would be to compute the RTG diameter on real datasets and compare it with measured generalization gaps; the paper only validates the diameter theorem on synthetic 2D grids.
  • The Hamming-1 edge rule may fail for deeper layers, where the zero sets of pre-activations are piecewise-linear rather than flat hyperplanes; if that happens, the correct structure may be a hypergraph or a stratified complex rather than a simple graph.
  • The compression theorem leaves the error constant $\delta$ unspecified in terms of region volumes; a quantitative bound relating $\delta$ to the volumes of pruned regions would make the result directly actionable for pruning algorithms.
  • The same graph construction could be applied to other piecewise-linear activation functions and to convolutional networks by treating each patch's activation pattern as a region, which would test whether the graph-theoretic approach generalizes beyond fully connected ReLU MLPs.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 5 minor

Summary. The paper introduces the ReLU Transition Graph (RTG), an undirected graph whose vertices are the linear regions (activation patterns) of a ReLU network and whose edges join regions whose activation patterns differ by exactly one bit. The main theoretical claims are: (Theorem 1) a bound on the number of RTG nodes; (Lemma 1) that Hamming-1 activation patterns correspond to facet-sharing regions; (Theorem 2) that the RTG is connected; (Lemma 2) that the region-volume entropy is at least log(d_avg+1); (Theorem 3) that the VC dimension of the network function class is at most the RTG diameter; (Lemma 3) a degree-sparsity statement; and (Theorem 4) a functional-compression guarantee by pruning low-degree RTG nodes. The paper also reports experiments on 2D synthetic data claiming to validate each result.

Significance. If valid, the RTG framework would provide a new graph-theoretic lens on ReLU expressivity, including a diameter-based upper bound on VC dimension and a topological justification for pruning. The paper also has the positive feature of attempting to validate every theoretical claim with reproducible experiments. However, the central capacity theorem is false, the validation of the core adjacency lemma is circular, and the compression theorem is not actually proved. Because the main advertised results do not withstand scrutiny and the defects are load-bearing rather than cosmetic, the manuscript in its current form does not meet the standard for publication.

major comments (5)
  1. [Section 4.5 and Appendix A.5] Theorem 3 is false as stated. Consider a two-hidden-unit ReLU network in R^2 with a linear output head. Its RTG has four nodes (0,0), (1,0), (0,1), (1,1) arranged in a 4-cycle, so diam(G)=2. The sign class shatters the three points (-1,-1), (1,-1), (-1,1): their activation codes (0,0), (1,0), (0,1) are affinely independent, so the linear output head can realize all eight labelings. Hence VC(F) >= 3 > 2 = diam(G). The proof's premise that shattering N points requires 2^N distinct activation patterns is invalid: within a single linear region the network is affine, and one affine function can shatter d+1 points without needing distinct activation patterns. The subsequent Sauer-type counting also fails in this regime: with m=2 and D=2 the Hamming-ball sum is 4 < 8, which would force VC <= 2, contradicting the example.
  2. [Section 4.3 and Appendix A.2] The load-bearing assumption of Theorem 2 and Lemma 1, stated in Section 4.3, is that 'each ReLU neuron partitions R^d via a non-degenerate hyperplane.' This holds only for first-layer neurons. For deeper neurons the pre-activation is piecewise linear and its zero set is a piecewise-linear hypersurface, not a single hyperplane. Consequently the proof of Lemma 1 in Appendix A.2, which treats H_k = {z_k(x)=0} as a hyperplane and concludes that two Hamming-1 patterns share a (d-1)-dimensional facet, does not apply to deep networks. Theorem 2's straight-line-path argument also assumes that every region crossing corresponds to a single neuron flip, which fails when the segment passes through intersections of several pre-activation boundaries. The connectivity and edge-addition rules therefore lack a valid foundation for the general networks the paper claims to analyze.
  3. [Section 5.3 and Section 5.3.2] The empirical validation of Lemma 1 is circular. Section 5.3 states that the RTG is constructed by 'connecting patterns with Hamming distance 1,' and Section 5.3.1 then reports that 100% of Hamming-1 pairs are connected. This is true by construction and does not test the geometric claim that such patterns correspond to facet-sharing regions. Similarly, Section 5.3.2 validates Lemma 2 by setting H(G)=log|V| and checking log|V| >= log(d_avg+1); this is not the entropy defined in Definition 4 (which is the Shannon entropy of the region-volume distribution), and the inequality log|V| >= log(d_avg+1) holds trivially for every simple graph because d_avg <= |V|-1. These experiments therefore provide no independent confirmation of the theorems.
  4. [Section 4.7 and Appendix A.7] Theorem 4 is not proved. Appendix A.7 does not specify the subnetwork f_core, the value of alpha in terms of d_avg, or the relationship between delta and the pruning set. The argument invokes unspecified 'volume-volume or degree-volume concentration arguments' and asserts without proof that low-degree regions have small volume. Moreover, the claimed uniform approximation error of arbitrarily small delta outside a measure-zero set cannot follow merely from removing low-degree nodes, since a low-degree region can still carry an arbitrarily large function value in an open set. The experimental protocol in Section 5.3.5 also does not test the theorem's quantified statement: it prunes 50% of nodes by degree and reports empirical max errors but never connects these to the claimed delta or to the measure-zero exceptional set.
  5. [Appendix A.1 and Section 4.1] Theorem 1's stated bound is not a valid upper bound. The proof correctly notes that one layer of width n induces at most sum_{i=0}^d C(n,i) regions, so a depth-L network has at most (sum_{i=0}^d C(n,i))^L regions. The appendix then claims the inequality (sum a_i)^L <= sum a_i^L, which is backwards for positive terms; e.g., (1+4)^2 = 25 > 1+16 = 17. Thus the final expression sum_{i=0}^d C(n,i)^L is smaller than the actual hyperplane-arrangement bound and cannot serve as a general upper bound. This is reflected in Table 1, where the 'theoretical upper bound' for L=2, n=4 is 53 while the correct product bound is 121.
minor comments (5)
  1. [Appendix A.1] The appendix heading says 'Existence of FSM-Emulating Neural Networks,' which does not match the content of the proof of Theorem 1.
  2. [Section 5.3.3] The text contains a typo: 'RTG diametere' should be 'RTG diameter.'
  3. [Section 5.3.2] The validation of Lemma 2 uses H(G)=log|V|, but Definition 4 defines H(G) as the Shannon entropy over region volumes; the reported quantity is not the entropy being theorized about.
  4. [Section 5.3.3] The VC-dimension 'proxy' used is the number of unique activation patterns on 10 random points, which is not a valid estimate of VC dimension and does not test the claimed inequality VC(F) <= diam(G).
  5. [Section 7] The limitations section acknowledges the low-dimensional and synthetic empirical setting, but it does not mention that the theoretical assumptions already fail outside the first layer; this should be disclosed.

Circularity Check

2 steps flagged · score 6.0 of 10

Partial circularity: the experimental 'validations' of Lemma 1 and Lemma 2 restate the graph-construction rules, while the main capacity theorem is mathematically unsound but not circular.

  1. self definitional [Section 5.3 'Validation of Lemma 1: Region Adjacency from Activation Distance' (RTG construction in Section 5.1)]
    "For each configuration, we extract all unique activation patterns and construct the RTG by placing edges between patterns that differ by exactly one bit (i.e., Hamming distance 1). ... we randomly sample 100 Hamming-1 activation pattern pairs and verify whether they are connected in the RTG. ... Hamming=1 pairs: 100/100 connected (100% correct) in all configurations and seeds"

    The experiment's RTG is built by the rule 'connect every Hamming-1 pair', so every sampled Hamming-1 pair is an edge by construction. The reported 100% connectedness is the construction rule restated, not evidence that such activation patterns share a (d-1)-dimensional boundary in input space. Lemma 1 is a geometric claim about facets; the validation only tests the graph-building code's implementation of its own edge rule.

  2. self definitional [Section 5.3.2 'Validation of Lemma 2: Entropy Grows with Average Degree']
    "we construct the RTG and compute the average degree of RTG nodes and the entropy H(G) = log|V| assuming a uniform region volume distribution. We find that in all cases, the entropy H(G) exceeds the theoretical bound log(davg + 1), validating Lemma 2."

    With the uniform-volume assumption, the entropy is set to H(G)=log|V| by Definition 4. But every simple graph satisfies d_avg+1 <= |V| (max degree <= |V|-1), so log(d_avg+1) <= log|V| is a universal graph inequality. The check cannot fail and is independent of ReLU geometry; it neither tests the volume-entropy claim nor distinguishes ReLU RTGs from random graphs. The validation reduces to a tautology.

full rationale

The paper's central mathematical derivation chain is not circular in the narrow sense: Theorems 2, 3, and 4 are asserted with proofs that do not fit parameters to data and do not rely on load-bearing self-citations. The self-citations [3] and [4] are only structural/style references, and no uniqueness theorem is imported from the authors' prior work. However, two 'empirical validations' are circular by construction. Lemma 1 is validated by checking that Hamming-1 pairs are connected in a graph whose edges were defined as exactly those pairs, so the test merely re-states the edge-construction rule. Lemma 2 is validated by assuming uniform region volumes, which makes H(G)=log|V| and reduces the claimed inequality to the universal graph fact d_avg+1 <= |V|. These two tautological checks are presented as confirmations of theoretical lemmas, giving partial circularity. In contrast, the serious mathematical defects in Theorem 3 (e.g., the Appendix A.5 claim that shattering N points requires 2^N distinct activation patterns, and the faulty Hamming-ball comparison) are correctness issues, not equivalence-to-input circularity, so they do not by themselves raise the circularity score further.

Assumptions & free parameters 0 free parameters · 5 assumptions · 1 invented entities

The framework loads four questionable premises: the hyperplane assumption for all neurons, the Hamming-1 adjacency rule, the uniform-volume entropy substitution, and the low-degree-implies-small-volume claim. None is independently justified, and the first two are false for deep networks.

assumptions (5)
  • ad hoc to paper Each ReLU neuron partitions R^d via a non-degenerate hyperplane.
    Invoked in Theorem 2 (Section 4.3) to justify connectivity and single-flip adjacency. False for hidden layers: pre-activations of neurons beyond the first layer are piecewise-linear, so their zero sets are not single hyperplanes.
  • ad hoc to paper Activation patterns differing in one bit always correspond to regions sharing a (d-1)-dimensional facet.
    The basis for edge construction (Definition 3 and Lemma 1). Fails for deep networks where a Hamming-1 pattern can be empty or adjacent across a boundary involving multiple neurons.
  • ad hoc to paper The number of distinct activation patterns over 10 random inputs is a valid proxy for VC dimension.
    Used in Section 5.3.3. The proxy is at most 10 and therefore trivially below any realistic RTG diameter; it does not measure shattering.
  • ad hoc to paper Low-degree RTG nodes have small total volume and can be removed with bounded functional error.
    Assumed in Theorem 4 (Appendix A.7). No bound on the union's measure is derived, so the 'arbitrarily small delta' claim is unsupported.
  • domain assumption Entropy can be computed as log|V| assuming uniform region volumes.
    Used in Section 5.3.2, deviating from the Definition 4 volume-based entropy. Under this assumption Lemma 2 collapses to a graph-theoretic tautology.
invented entities (1)
  • ReLU Transition Graph (RTG)
    purpose: A graph over linear regions to analyze connectivity, VC dimension, and compression.
    The object is defined in the paper, but no falsifiable prediction outside the paper is attached to it, and the experiments used to support it test the construction rule itself.

how reviews work

0 comments
Cite this review

Pith. "Pith review of The Geometry of ReLU Networks through the ReLU Transition Graph." pith.science (2026). https://pith.science/paper/6D5LXQU7

@misc{pith2026250511692,
  author       = {Pith},
  title        = {Pith review of: The Geometry of ReLU Networks through the ReLU Transition Graph},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6D5LXQU7}},
  note         = {Machine review of arXiv:2505.11692}
}
read the original abstract

We develop a novel theoretical framework for analyzing ReLU neural networks through the lens of a combinatorial object we term the ReLU Transition Graph (RTG). In this graph, each node corresponds to a linear region induced by the network's activation patterns, and edges connect regions that differ by a single neuron flip. Building on this structure, we derive a suite of new theoretical results connecting RTG geometry to expressivity, generalization, and robustness. Our contributions include tight combinatorial bounds on RTG size and diameter, a proof of RTG connectivity, and graph-theoretic interpretations of VC-dimension. We also relate entropy and average degree of the RTG to generalization error. Each theoretical result is rigorously validated via carefully controlled experiments across varied network depths, widths, and data regimes. This work provides the first unified treatment of ReLU network structure via graph theory and opens new avenues for compression, regularization, and complexity control rooted in RTG analysis.

Figures

Figures reproduced from arXiv: 2505.11692 by the authors.

Figure 1
Figure 1. Empirical RTG node counts vs. theoretical upper bound for various network depths [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Validation of Lemma 2. Each point corresponds to a network configuration. All lie above [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. VC-dimension proxy vs. RTG diameter across [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Validation of Theorem 4. Each point shows approximation error vs. compression rate after [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Discrete Functional Geometry of ReLU Networks via ReLU Transition Graphs

    cs.LG 2025-09 reject novelty 4.0 of 10

    The paper claims ReLU Transition Graphs of ReLU networks are expanders whose spectral gap, region entropy, and edge KL divergence bound generalization and capacity; the proofs are sketches, and the empirical checks are weak.

Reference graph

Works this paper leans on

23 extracted references · 15 canonical work pages · cited by 1 Pith paper

  1. [1]

    Understanding deep neural networks with rectified linear units, 2018

    Raman Arora, Amitabh Basu, Poorya Mianjy, and Anirbit Mukherjee. Understanding deep neural networks with rectified linear units, 2018

    2018

  2. [2]

    Bartlett

    P.L. Bartlett. The sample complexity of pattern classification with neural networks: the size of the weights is more important than the size of the network. IEEE Transactions on Information Theory, 44(2):525–536, 1998

    work page 1998

  3. [3]

    A combinatorial theory of dropout: Subnetworks, graph geometry, and generalization, 2025

    Sahil Rajesh Dhayalkar. A combinatorial theory of dropout: Subnetworks, graph geometry, and generalization, 2025

    2025

  4. [4]

    Neural networks as universal finite-state machines: A constructive deterministic finite automaton theory, 2025

    Sahil Rajesh Dhayalkar. Neural networks as universal finite-state machines: A constructive deterministic finite automaton theory, 2025

    work page 2025

  5. [5]

    The lottery ticket hypothesis: Finding sparse, trainable neural networks, 2019

    Jonathan Frankle and Michael Carbin. The lottery ticket hypothesis: Finding sparse, trainable neural networks, 2019

    2019

  6. [6]

    Deep sparse rectifier neural networks

    Xavier Glorot, Antoine Bordes, and Yoshua Bengio. Deep sparse rectifier neural networks. In Geoffrey Gordon, David Dunson, and Miroslav Dudík, editors, Proceedings of the Fourteenth International Conference on Artificial Intelligence and Statistics, volume 15 of Proceedings of Machine Learning Research, pages 315–323, Fort Lauderdale, FL, USA, 11–13 Apr 2...

    work page 2011

  7. [7]

    Guss and Ruslan Salakhutdinov

    William H. Guss and Ruslan Salakhutdinov. On characterizing the capacity of neural networks using algebraic topology, 2018

    2018

  8. [8]

    Complexity of linear regions in deep networks

    Boris Hanin and David Rolnick. Complexity of linear regions in deep networks. In Kamalika Chaudhuri and Ruslan Salakhutdinov, editors,Proceedings of the 36th International Conference on Machine Learning, volume 97 of Proceedings of Machine Learning Research, pages 2596–

    work page

Show all 23 references
  1. [9]

    Approximating continuous functions by relu nets of minimal width, 2018

    Boris Hanin and Mark Sellke. Approximating continuous functions by relu nets of minimal width, 2018. 9

    2018

  2. [10]

    On the number of linear regions of deep neural networks

    Guido F Montúfar, Razvan Pascanu, Kyunghyun Cho, and Yoshua Bengio. On the number of linear regions of deep neural networks. In NeurIPS, 2014

    2014

  3. [11]

    Vinod Nair and Geoffrey E. Hinton. Rectified linear units improve restricted boltzmann machines. In Proceedings of the 27th International Conference on International Conference on Machine Learning, ICML’10, page 807–814, Madison, WI, USA, 2010. Omnipress

    2010

  4. [12]

    In search of the real inductive bias: On the role of implicit regularization in deep learning, 2015

    Behnam Neyshabur, Ryota Tomioka, and Nathan Srebro. In search of the real inductive bias: On the role of implicit regularization in deep learning, 2015

    2015

  5. [13]

    Norm-based capacity control in neural networks

    Behnam Neyshabur, Ryota Tomioka, and Nathan Srebro. Norm-based capacity control in neural networks. In Peter Grünwald, Elad Hazan, and Satyen Kale, editors, Proceedings of The 28th Conference on Learning Theory, volume 40 of Proceedings of Machine Learning Research, pages 1376...

    2015

  6. [14]

    Abolafia, Jeffrey Pennington, and Jascha Sohl- Dickstein

    Roman Novak, Yasaman Bahri, Daniel A. Abolafia, Jeffrey Pennington, and Jascha Sohl- Dickstein. Sensitivity and generalization in neural networks: an empirical study, 2018

    2018

  7. [15]

    Pytorch: An imperative style, high- performance deep learning library

    Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, Alban Desmaison, Andreas Kopf, Edward Yang, Zachary DeVito, Martin Raison, Alykhan Tejani, Sasank Chilamkurthy, Benoit Steiner, Lu ...

    2019

  8. [16]

    Expo- nential expressivity in deep neural networks through transient chaos

    Ben Poole, Subhaneil Lahiri, Maithra Raghu, Jascha Sohl-Dickstein, and Surya Ganguli. Expo- nential expressivity in deep neural networks through transient chaos. In D. Lee, M. Sugiyama, U. Luxburg, I. Guyon, and R. Garnett, editors, Advances in Neural Information Processing Sy...

    2016

  9. [17]

    On the expressive power of deep neural networks, 2017

    Maithra Raghu, Ben Poole, Jon Kleinberg, Surya Ganguli, and Jascha Sohl-Dickstein. On the expressive power of deep neural networks, 2017

    2017

  10. [18]

    Bounding and counting linear regions of deep neural networks

    Thiago Serra, Christian Tjandraatmadja, and Srikumar Ramalingam. Bounding and counting linear regions of deep neural networks. In Jennifer Dy and Andreas Krause, editors,Proceedings of the 35th International Conference on Machine Learning , volume 80 of Proceedings of Machine ...

    2018

  11. [19]

    benefits of depth in neural networks

    Matus Telgarsky. benefits of depth in neural networks. In Vitaly Feldman, Alexander Rakhlin, and Ohad Shamir, editors, 29th Annual Conference on Learning Theory, volume 49 of Proceed- ings of Machine Learning Research, pages 1517–1539, Columbia University, New York, New York, ...

    2016

  12. [20]

    Facing up to arrangements: face-count formulas for partitions of space by hyperplanes, volume 1

    Thomas Zaslavsky. Facing up to arrangements: face-count formulas for partitions of space by hyperplanes, volume 1. 1975

    1975

  13. [21]

    Lee, Martin J

    Yuchen Zhang, Jason D. Lee, Martin J. Wainwright, and Michael I. Jordan. Learning halfspaces and neural networks with random initialization, 2015. A Appendix A.1 Proof of Theorem 1: Existence of FSM-Emulating Neural Networks The proof follows a classical result from the theory...

    2015

  14. [23]

    For each such nodev∈ S, the corresponding region Rv contributes little to the function’s global variation due to its low connectivity (few adjacent regions) and likely small volume

    such that eachv∈S has low degree. For each such nodev∈ S, the corresponding region Rv contributes little to the function’s global variation due to its low connectivity (few adjacent regions) and likely small volume. Define: fcore(x) := f (x)· ⊮[x /∈∪ v∈SRv] + ˜f (x)· ⊮[x∈∪ v∈S...

  15. [2604]

    PMLR, 09–15 Jun 2019

    2019

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.