Pith. sign in

REVIEW 4 major objections 6 minor 39 references

A training-free, forward-only probe of hidden-state sensitivity can assign LoRA ranks before fine-tuning and beat uniform allocation on all three tested model–benchmark settings.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-04 04:28 UTC pith:OXL5LF52

load-bearing objection A genuinely new allocation heuristic with a plausible mechanism, but the empirical case as reported is undermined by implausible baselines and a ~40% parameter-budget mismatch. the 4 major comments →

arxiv 2607.09757 v2 pith:OXL5LF52 submitted 2026-07-05 cs.CV cs.AI

RSRA: Training-Free Probing of Representation Sensitivity for Efficient LoRA Rank Allocation

Jiaqi Liu , Haidong Kang , Qihui Zhao , Guo Yu , Jingchao Wang This is my paper
classification cs.CV cs.AI
keywords LoRA rank allocationrepresentation sensitivityFréchet distancespectral effective ranktraining-free probingparameter-efficient fine-tuningcalibration setforward-only estimation
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper claims that where a Low-Rank Adaptation (LoRA) budget should be spent can be decided before any training, by probing how much each module's hidden-state distribution would move under a standardized random low-rank update. RSRA combines two forward-only signals — the spectral breadth of each layer's activations and the Fréchet-distance displacement caused by a virtual probe on each module — into a hierarchical rank allocation. If correct, this replaces training-time rank pruning and gradient-based pre-allocation with a cheap calibration pass and generalizes across adapter variants. The authors report the highest average accuracy in every model–benchmark setting they tested (Qwen3-4B and Mistral-7B on commonsense reasoning; Mistral-7B on GLUE), a 1.48–1.93x speedup in allocation time over the fastest competing pre-allocation method, and gains on 15 of 18 combinations when plugged into DoRA, LoRA-FA, and PiSSA.

Core claim

RSRA asserts that representation sensitivity — the change in a layer's hidden-state distribution caused by a module-specific low-rank update — can be measured before fine-tuning with forward passes only. On a 256-sample calibration set, it computes each layer's Spectral Effective Rank from the singular values of its centered activations, then applies a standardized virtual low-rank update to each module (random Gaussian factors scaled to epsilon=0.01 of the weight's Frobenius norm). The squared Fréchet distance between baseline and probed layer statistics, normalized by baseline energy, gives an intra-layer sensitivity score; hierarchical normalization combines it with the spectral share to

What carries the argument

The central object is the standardized virtual low-rank probe (Eq. 6): a random Gaussian low-rank matrix, normalized to a fixed relative magnitude epsilon=0.01 of the target weight's Frobenius norm, applied to one module at a time. The sensitivity score (Eq. 8) is the squared Fréchet distance between the baseline layer hidden-state distribution and the probed distribution, normalized by the baseline's mean and covariance energy. This intra-layer signal is combined with the layer's Spectral Effective Rank (Eq. 5) through hierarchical normalization (Eq. 9), and the resulting global importance scores are projected onto integer ranks in a bounded range (Eq. 11). The Fréchet distance matters beca

Load-bearing premise

The method assumes that one fixed probe magnitude (epsilon=0.01, chosen from trained LoRA update sizes on three GLUE tasks) and a 256-sample calibration set capture, before training, the same module importance that emerges after training.

What would settle it

Run RSRA on a task whose trained LoRA update magnitudes lie well outside the 0.004–0.018 range used to set epsilon, and compare its allocation against a random permutation of the same rank budget; if the RSRA allocation does not beat the random one in downstream accuracy, the fixed-epsilon probe is not task-relevant.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Rank allocation can be fully decoupled from fine-tuning: the rank configuration is computed once, before training, so no iterative pruning or dynamic budget adjustment is needed during optimization.
  • The allocation signal transfers across adapter variants (DoRA, LoRA-FA, PiSSA) because it is independent of parameterization, initialization, and optimizer, making RSRA a drop-in allocator for other PEFT methods.
  • The sensitivity ranking is stable with small calibration sets — the paper reports high Spearman correlations even at 64 samples — so the probe's cost stays low for large models.
  • On harder tasks (e.g., ARC-Challenge), a non-zero minimum rank per module is important, meaning allocation should respect rank floors rather than aggressively sparsifying modules.
  • Allocation time grows only mildly with calibration size (a 16x increase in samples produces about a 1.73x increase in time), so practitioners can trade a little accuracy for speed by shrinking the calibration set.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The probe uses random directions, not task-optimized ones. A testable extension would compare random probes with gradient-informed or data-adaptive probe directions; if random directions suffice, the signal is carried by the geometry of the representation space itself, not by the specific update content.
  • The fixed epsilon=0.01 was calibrated on three GLUE tasks (MRPC, RTE, WNLI) that later appear in the evaluation. A robustness check would measure trained update magnitudes on a held-out task family and verify that re-running RSRA with those magnitudes keeps the rank configuration nearly unchanged.
  • The Fréchet-distance covariance term may be noisy on small calibration sets. A mean-only displacement variant (or a different distribution distance) would isolate which component of the sensitivity score carries the predictive signal.
  • Since the probe is forward-only, it could also be used to freeze low-sensitivity modules entirely (rank 0) rather than just giving them fewer dimensions, which would further reduce trainable parameters without harming performance.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. RSRA is a forward-only, pre-fine-tuning rank allocator for LoRA. It scores layers by Spectral Effective Rank and modules by the Fréchet-distance displacement of hidden-state statistics under a standardized random low-rank probe, then combines these scores by hierarchical normalization into a rank configuration with a prescribed average rank. The paper reports experiments on Qwen3-4B and Mistral-7B across commonsense reasoning and GLUE, claiming the highest average accuracy in all three model–benchmark settings, a 1.48×–1.93× allocation-time speedup over the fastest competing pre-allocation method, and consistent gains when integrated with DoRA, LoRA-FA, and PiSSA.

Significance. If the empirical claims were reliable, RSRA would be a practically valuable pre-allocation method: it is simple, forward-only, potentially model-agnostic, and the paper includes useful ablations of calibration-set size and rank bounds. The motivation from heterogeneous module sensitivity is clearly presented. However, the central comparison is not currently trustworthy: the paper's own efficiency tables contradict the stated equal-parameter-budget condition, several published baselines are reported at implausibly low accuracy, and the probe magnitude is calibrated on tasks that appear in the evaluation suite. The core contribution is therefore not established by the present experiments.

major comments (4)
  1. [Appendix B Table 6 vs Appendix C Table 8] The paper states that the average rank is maintained at r=8 to ensure a fair parameter-budget comparison with LoRA (Table 6), but Table 8 reports trainable parameters of 41.94M for LoRA versus 25.35M for RSRA on PIQA and 41.94M versus 25.61M on CoLA, a ~40% reduction. The appendix even acknowledges that RSRA reduces the parameter count by approximately 40% compared with uniform LoRA. Thus the headline accuracy comparisons against LoRA and AdaLoRA are not made under the claimed equal-budget condition, and the gains could arise from implicit regularization or reduced capacity rather than from the sensitivity-based allocation.
  2. [Tables 1, 2, and 4] Several baseline results are implausible for published methods. GoRA drops to 52.77 average on Mistral-7B commonsense (e.g., BoolQ 47.49 vs LoRA 89.42); PiSSA baselines in Table 4 are 48.38 on QNLI, 40.33 on QQP, and 22.95 on ARC-C; AdaLoRA underperforms vanilla LoRA by large margins on many tasks. These numbers strongly suggest misconfigured hyperparameters or implementation errors for the baselines. Consequently the aggregate 'improves 15 of 18 combinations' claim and the claimed advantage over competing allocators are not credible until the baselines are verified.
  3. [Appendix C, Eq. (6), Table 9] The probe magnitude epsilon=0.01 is set by measuring the relative Frobenius norm of trained LoRA updates on MRPC, RTE, and WNLI (Table 9). These three tasks are part of the GLUE evaluation in Table 2. The probe magnitude is therefore calibrated on a subset of the evaluation target, which undermines the claim that epsilon is a task-agnostic, conservative choice. The authors should calibrate epsilon on held-out tasks (or show insensitivity across a broader task set that excludes the GLUE test tasks) before claiming that the same epsilon applies universally.
  4. [Section 4, RQ4, Figure 5] The validation of the sensitivity signal is confounded by the rank allocation itself. Modules are ranked by RSRA sensitivity and the top/middle/bottom 10% groups are ablated after fine-tuning with RSRA's rank configuration. Because high-sensitivity modules are assigned larger ranks, their trained updates contain more parameters and hence more capacity; the larger loss increase after ablation could reflect rank/capacity rather than the intrinsic relevance of the sensitivity score. To support the claimed alignment, the ablation should be run under a uniform-rank configuration or the contribution should be normalized by capacity.
minor comments (6)
  1. [Appendix C, Table 7] The abstract says RSRA achieves the highest average performance in 'all three reported model–benchmark settings,' but the mathematical-reasoning experiments in Table 7 show AdaLoRA outperforming RSRA on average (40.04 vs 38.44). Please qualify the claim or explicitly exclude this setting.
  2. [Table 4] Unlike Tables 1 and 2, Table 4 reports no standard deviations or number of seeds. Given the very large PiSSA baseline swings, this table needs error bars and a description of how many seeds were used.
  3. [Section 3 and Appendix B] RSRA has several free parameters (epsilon, r_probe, N, Rmin, Rmax). The paper ablates N and Rmin/Rmax, but not r_probe or epsilon (except for the GLUE-calibrated choice). Please report sensitivity to r_probe and epsilon on a held-out task.
  4. [Eq. (9)] The global importance I_{l,j} is a product of two normalized scores. A product is not the only way to combine inter-layer and intra-layer signals; the paper should justify why multiplication is preferable to addition or another aggregation, especially since the product's scale affects rounding in Eq. (11).
  5. [General] The term 'training-free' is potentially misleading: RSRA requires a task-specific calibration set and many forward passes. Please clarify that 'training-free' means no gradient-based optimization during allocation, not that the procedure is free of data or compute.
  6. [General] No code is provided. Given the sensitivity of the results to hyperparameter configuration and the implausible baselines, releasing code and exact baseline configurations is essential for reproducibility.

Circularity Check

0 steps flagged

No significant circularity: the only calibration-like choice (probe scale epsilon) cancels in the within-layer normalization, and no prediction is defined in terms of its own inputs.

full rationale

RSRA is an empirical rank-allocation heuristic rather than a derivation whose outputs are defined in terms of its inputs. The one data-dependent choice that could look like fitting is the probe magnitude epsilon=0.01, set in Appendix C from trained LoRA update magnitudes on MRPC/RTE/WNLI, which also appear in the GLUE evaluation. However, Eq. (8) computes s_{l,j} as a squared Frechet displacement normalized by a baseline quantity independent of the probe, and Eq. (9) uses s_{l,j} only through the intra-layer ratio s_{l,j}/sum_m s_{l,m}. Since the same epsilon multiplies every probe update, a common factor epsilon^2 cancels in that ratio (and the inter-layer effective-rank factor does not involve epsilon). The rank configuration of Eq. (11) is therefore not determined by the value of epsilon in the small-perturbation regime, so calibrating epsilon on part of the evaluation suite does not force the reported GLUE results. The parameter-count discrepancy between Table 8 (RSRA ~25.4M params on PIQA) and the stated 'average rank r=8' fair-budget condition is an internal numerical inconsistency bearing on experimental fairness, not a circularity in the derivation. RQ4's top/bottom ablation is confounded by the fact that top-sensitivity modules receive larger allocated ranks, so removing their trained updates removes more parameters; this is a validity concern, but it does not make the method's predictions equivalent to its inputs by construction. No load-bearing self-citations, uniqueness theorems, or ansatz-smuggling citations are used: effective rank and Frechet distance are external definitions, and all evaluations are against external benchmarks. The central claim rests on empirical comparisons, not on a chain that reduces to its own inputs.

Axiom & Free-Parameter Ledger

5 free parameters · 5 axioms · 0 invented entities

RSRA introduces no new physical entities. Its main burden is a set of modeling choices: spectral effective rank as a capacity proxy, Fréchet displacement as sensitivity, a probe scale fitted on evaluation tasks, and a heuristic normalization for combining signals.

free parameters (5)
  • epsilon (probe magnitude) = 0.01
    Relative Frobenius norm of the virtual update; chosen to match trained LoRA update magnitudes measured on MRPC/RTE/WNLI (Appendix C, Table 9), which are themselves GLUE evaluation tasks.
  • r_probe = 8
    Virtual probe rank; set by hand, not ablated in the paper.
  • calibration size N = 256
    Calibration set size; selected after stability ablation (Table 10), not derived from first principles.
  • Rmin = 2
    Lower rank clamp; chosen from ablation experiments (Tables 11–12).
  • Rmax = 128
    Upper rank clamp; chosen from ablation experiments (Tables 11–12).
axioms (5)
  • domain assumption Spectral effective rank of centered activations is a valid proxy for how much LoRA capacity a layer needs
    Invoked in Eq. (5); no theoretical link is given between spectral breadth of pretrained features and LoRA update requirements.
  • domain assumption Fréchet distance under a random Gaussian low-rank probe measures task-relevant sensitivity
    Eqs. (6)–(8); validated only post hoc in RQ4, not derived from adaptation dynamics.
  • domain assumption 256 calibration samples are representative of the task's representation geometry
    Used to estimate Eq. (1); no coverage or concentration guarantee is provided.
  • ad hoc to paper The same epsilon=0.01 is valid for all layers, modules, and tasks
    Appendix C's own Table 9 reports trained update magnitudes ranging from 0.00453 to 0.01805, so a single fixed value is a simplification.
  • ad hoc to paper Hierarchical normalization (Eq. 9) is a sensible way to combine the two signals
    The product of normalized spectral rank and normalized Fréchet score is asserted; no ablation compares it to alternatives.

pith-pipeline@v1.3.0-alltime-deepseek · 138 in / 14983 out tokens · 165388 ms · 2026-08-04T04:28:13.937699+00:00 · methodology

0 comments
read the original abstract

Parameter-efficient fine-tuning enables large language models to adapt to downstream tasks with substantially lower computational and storage cost, and Low-Rank Adaptation (LoRA) is among its most widely used techniques. However, vanilla LoRA assigns a uniform rank to all adapted modules, while existing adaptive methods either incur additional optimization overhead or rely on static weights and local gradients that do not capture task-conditioned representation changes. We propose RSRA, a training-free rank allocator that estimates where adaptation capacity is most needed through forward-only representation sensitivity probing on a small calibration set. Specifically, RSRA uses Spectral Effective Rank to allocate capacity across layers, measures module-wise hidden-state displacement under standardized virtual low-rank updates with the Frechet Distance, and combines both signals through hierarchical normalization to produce a task-aware rank configuration before fine-tuning. Across commonsense reasoning and natural language understanding benchmarks with Qwen3-4B and Mistral-7B, RSRA achieves the highest average performance in all three reported model-benchmark settings and a 1.48x-1.93x speedup in allocation time over the fastest competing pre-allocation method. When integrated with DoRA, LoRA-FA, and PiSSA, RSRA improves 15 of the 18 evaluated combinations and increases the average performance of all three PEFT methods.

Figures

Figures reproduced from arXiv: 2607.09757 by Guo Yu, Haidong Kang, Jiaqi Liu, Jingchao Wang, Qihui Zhao.

Figure 1
Figure 1. Figure 1: Layer-wise representational sensitivity induced by module-specific adaptation, measured [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 1
Figure 1. Figure 1: Performance–efficiency trade-off on GLUE with [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Computation and parameter efficiency analysis. The bars represent the extra training [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 2
Figure 2. Figure 2: Heatmap of representation-sensitivity percentiles [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Loss increase on BoolQ after removing differ [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Efficiency comparison on (a) PIQA and (b) CoLA. [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Validation of the representation-sensitivity scores [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗

Review history (2 revisions) →

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

39 extracted references · 12 linked inside Pith

  1. [1]

    Advances in neural information processing systems , volume=

    Language models are few-shot learners , author=. Advances in neural information processing systems , volume=

  2. [2]

    arXiv preprint arXiv:2203.02155 , year=

    Training language models to follow instructions with human feedback , author=. arXiv preprint arXiv:2203.02155 , year=

    Pith/arXiv arXiv

  3. [3]

    International conference on machine learning , pages=

    Parameter-efficient transfer learning for NLP , author=. International conference on machine learning , pages=. 2019 , organization=

    2019

  4. [4]

    arXiv preprint arXiv:2110.04366 , year=

    Towards a unified view of parameter-efficient transfer learning , author=. arXiv preprint arXiv:2110.04366 , year=

    Pith/arXiv arXiv

  5. [5]

    , author=

    Lora: Low-rank adaptation of large language models. , author=

  6. [6]

    arXiv preprint arXiv:2303.10512 , year=

    Adalora: Adaptive budget allocation for parameter-efficient fine-tuning , author=. arXiv preprint arXiv:2303.10512 , year=

    Pith/arXiv arXiv

  7. [7]

    AIRA: Activation-Informed Low-Rank Adaptation for Large Models , author=

  8. [8]

    Sparse low-rank adaptation of pre-trained language models , author=

  9. [9]

    Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics , pages=

    DyLoRA: Parameter-efficient tuning of pre-trained models using dynamic search-free low-rank adaptation , author=. Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics , pages=

  10. [10]

    Alora: Allocating low-rank adaptation for fine-tuning large language models , author=

  11. [11]

    , author=

    Beyond Low-Rank Tuning: Model Prior-Guided Rank Allocation for Effective Transfer in Low-Data and Large-Gap Regimes. , author=

  12. [12]

    arXiv preprint arXiv:2502.12171 , year=

    Gora: Gradient-driven adaptive low rank adaptation , author=. arXiv preprint arXiv:2502.12171 , year=

    arXiv

  13. [13]

    The journal of machine learning research , volume=

    A kernel two-sample test , author=. The journal of machine learning research , volume=. 2012 , publisher=

    2012

  14. [14]

    Dowson, David C and Landau, BV666017 , journal=. The Fr. 1982 , publisher=

    1982

  15. [15]

    Expositiones mathematicae , volume=

    On the Bures--Wasserstein distance between positive definite matrices , author=. Expositiones mathematicae , volume=. 2019 , publisher=

    2019

  16. [16]

    The Thirteenth International Conference on Learning Representations , year=

    HiRA: Parameter-efficient hadamard high-rank adaptation for large language models , author=. The Thirteenth International Conference on Learning Representations , year=

  17. [17]

    Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers) , pages=

    Autolora: Automatically tuning matrix ranks in low-rank adaptation based on meta learning , author=. Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers) , pages=

    2024

  18. [18]

    arXiv preprint arXiv:2310.11454 , year=

    Vera: Vector-based random matrix adaptation , author=. arXiv preprint arXiv:2310.11454 , year=

    Pith/arXiv arXiv

  19. [19]

    arXiv preprint arXiv:2505.20355 , year=

    Gralora: Granular low-rank adaptation for parameter-efficient fine-tuning , author=. arXiv preprint arXiv:2505.20355 , year=

    arXiv

  20. [20]

    Boolq: Exploring the surprising difficulty of natural yes/no questions , author=. Proceedings of the 2019 conference of the north American chapter of the association for computational linguistics: Human language technologies, volume 1 (long and short papers) , pages=

    2019

  21. [21]

    Proceedings of the AAAI conference on artificial intelligence , volume=

    Piqa: Reasoning about physical commonsense in natural language , author=. Proceedings of the AAAI conference on artificial intelligence , volume=

  22. [22]

    Social IQa: Commonsense reasoning about social interactions , author=. Proceedings of the 2019 conference on empirical methods in natural language processing and the 9th international joint conference on natural language processing (EMNLP-IJCNLP) , pages=

    2019

  23. [23]

    Proceedings of the 57th annual meeting of the association for computational linguistics , pages=

    Hellaswag: Can a machine really finish your sentence? , author=. Proceedings of the 57th annual meeting of the association for computational linguistics , pages=

  24. [24]

    Communications of the ACM , volume=

    Winogrande: An adversarial winograd schema challenge at scale , author=. Communications of the ACM , volume=. 2021 , publisher=

    2021

  25. [25]

    arXiv preprint arXiv:1803.05457 , year=

    Think you have solved question answering? try arc, the ai2 reasoning challenge , author=. arXiv preprint arXiv:1803.05457 , year=

    Pith/arXiv arXiv

  26. [26]

    Proceedings of the 2018 conference on empirical methods in natural language processing , pages=

    Can a suit of armor conduct electricity? a new dataset for open book question answering , author=. Proceedings of the 2018 conference on empirical methods in natural language processing , pages=

    2018

  27. [27]

    Zenodo , year=

    A framework for few-shot language model evaluation , author=. Zenodo , year=

  28. [28]

    arXiv preprint arXiv:2505.09388 , year=

    Qwen3 technical report , author=. arXiv preprint arXiv:2505.09388 , year=

    Pith/arXiv arXiv

  29. [29]

    arXiv preprint arXiv:2310.06825 , year =

    Mistral 7B , author =. arXiv preprint arXiv:2310.06825 , year =

    Pith/arXiv arXiv

  30. [30]

    arXiv preprint arXiv:2501.12948 , year=

    Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning , author=. arXiv preprint arXiv:2501.12948 , year=

    Pith/arXiv arXiv

  31. [31]

    arXiv preprint arXiv:2103.03874 , year=

    Measuring mathematical problem solving with the math dataset , author=. arXiv preprint arXiv:2103.03874 , year=

    Pith/arXiv arXiv

  32. [32]

    arXiv preprint arXiv:2110.14168 , year=

    Training verifiers to solve math word problems , author=. arXiv preprint arXiv:2110.14168 , year=

    Pith/arXiv arXiv

  33. [33]

    Proceedings of the 2018 EMNLP workshop BlackboxNLP: Analyzing and interpreting neural networks for NLP , pages=

    GLUE: A multi-task benchmark and analysis platform for natural language understanding , author=. Proceedings of the 2018 EMNLP workshop BlackboxNLP: Analyzing and interpreting neural networks for NLP , pages=

    2018

  34. [34]

    arXiv preprint arXiv:2312.03732 , year=

    A rank stabilization scaling factor for fine-tuning with lora , author=. arXiv preprint arXiv:2312.03732 , year=

    Pith/arXiv arXiv

  35. [35]

    Dora: Weight-decomposed low-rank adaptation , author=

  36. [36]

    arXiv preprint arXiv:2308.03303 , year=

    Lora-fa: Memory-efficient low-rank adaptation for large language models fine-tuning , author=. arXiv preprint arXiv:2308.03303 , year=

    Pith/arXiv arXiv

  37. [37]

    Advances in Neural Information Processing Systems , volume=

    Pissa: Principal singular values and singular vectors adaptation of large language models , author=. Advances in Neural Information Processing Systems , volume=

  38. [38]

    2007 15th European signal processing conference , pages=

    The effective rank: A measure of effective dimensionality , author=. 2007 15th European signal processing conference , pages=. 2007 , organization=

    2007

  39. [39]

    Advances in neural information processing systems , volume=

    Are gans created equal? a large-scale study , author=. Advances in neural information processing systems , volume=