Pith. sign in

REVIEW 4 major objections 6 minor 41 references

Binary and Ternary Quantization Can Enhance Feature Discrimination

T0 review · 4 major / 6 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read Coarse 1-bit and 2-bit quantization can sharpen class separation instead of degrading it, provided the threshold is chosen correctly.

desk verdict The scalar theory is solid and worth reviewing, but the paper's vector-level claim rests on an unproved aggregation step and overstates what the experiments show. read the letter →

arxiv 2504.13792 v2 pith:EQPWZLFY submitted 2025-04-18 cs.LG

Weizhi Lu , Mingrui Chen , Weiyu Li This is my paper · ORCID
classification cs.LG MSC 62H30
keywords binaryquantizationternaryfeaturediscriminationclassificationaccuracythresholdGaussianmixturemodelFisherlineardiscriminanterror
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 sets out to explain an empirical oddity that conventional quantization theory dismisses as an accident: replacing real-valued features with 1-bit {0,1} or 2-bit {0,±1} codes sometimes improves classification accuracy instead of destroying it. The proposal is to judge a quantizer by feature discrimination — the ratio of between-class spread to within-class spread — rather than by reconstruction error, the usual yardstick. Under a two-class Gaussian model with equal variances and standardized means, the paper proves that a binary or ternary threshold exists at which the quantized discrimination exceeds the original, and it pins down the enhancing threshold ranges with explicit inequalities. Classification experiments on synthetic data and on image, speech, and text features show accuracy at those thresholds reaching or beating full-precision accuracy. If the proof holds, the practical message is that aggressive quantization is not merely a tolerable loss: the threshold itself is a tunable lever that can help the downstream classifier.

What carries the argument

The load-bearing object is the discrimination ratio $D = E[(X_1-Y_1)^2] / ( E[(X_1-X_2)^2] + E[(Y_1-Y_2)^2] )$, a linear-discrimination quotient of inter-class to intra-class scatter, together with its quantized counterparts $D_b$ and $D_t$. Under the standardized Gaussian model of Property 1, the quantized moments collapse to polynomials in the standard-normal tail probabilities $\alpha = \Phi((\tau-\mu)/\sigma)$ and $\beta = \Phi((\tau+\mu)/\sigma)$, so the comparison $D_b > D$ or $D_t > D$ becomes a quadratic inequality in $\alpha$. Solving the quadratic and discarding the root that conflicts with $\alpha < \beta$ yields conditions (8) and (9), whose left-hand sides can be evaluated or minimized over $\tau$. This reduction is what turns a nonlinear quantizer into a solvable threshold-selection problem, and the same machinery supplies the gradient-descent recipe of Appendix B for locating enhancing thresholds.

What would settle it

Take a real feature set with strong inter-coordinate correlations, such as raw image patches or unwhitened speech spectra, run the paper's threshold search on it, and test whether any threshold yields quantized accuracy above the full-precision baseline on held-out data; if none does, the per-coordinate independence assumption, rather than quantization itself, is carrying the effect.

Watch

Extended reading notes

Core claim

Modeling two classes as Gaussian variables $X \sim N(\mu, \sigma^2)$ and $Y \sim N(-\mu, \sigma^2)$ that have been standardized so that $\mu^2 + \sigma^2 = 1$ (Property 1), the paper defines discrimination as $D = E[(X_1-Y_1)^2] / ( E[(X_1-X_2)^2] + E[(Y_1-Y_2)^2] )$, the quotient of inter-class to intra-class squared distance. Its central results are Theorem 4 and Theorem 5: binary quantization at threshold $\tau$ satisfies $D_b > D$ whenever inequality (8) holds, and ternary quantization satisfies $D_t > D$ whenever inequality (9) holds, where the two inequalities are algebraic conditions in $\mu$, $\sigma$, and the standard-normal tail probabilities $\alpha$ and $\beta$ evaluated at the shifted thresholds. The same inequalities double as a recipe for finding the enhancing threshold by gradient descent. The authors show numerically that such thresholds exist when classes are sufficiently separated — $\mu$ above roughly 0.76 for binary and 0.66 for ternary — and that at those thresholds, KNN, SVM, MLP, and decision-tree classifiers on synthetic and real data reach or exceed full-precision accuracy. Their intended conclusion is that feature discrimination, not quantization error, is the quantity that predicts whether coarse quantization helps classification.

Load-bearing premise

The theory assumes each coordinate of a feature vector is Gaussian with the same variance in both classes and is nearly independent of the other coordinates, so strongly correlated real features or unequal class variances can invalidate the threshold conditions.

Editorial extensions

If this is right

  • For classes that are well separated, there is a whole interval of thresholds — not a single lucky value — at which 1-bit or 2-bit quantized features outperform full-precision features, so threshold placement becomes a first-class design step in quantization pipelines.
  • Ternary quantization offers more opportunities than binary: enhancing thresholds exist for less separated classes ($\mu$ above about 0.66, versus 0.76 for binary) and over a broader interval of $\tau$.
  • The $\{0,1\}$ binary result carries over to the widely used $\{-1,1\}$ binary encoding, because Euclidean distance on the former equals cosine distance on the latter.
  • Because discrimination is analyzed per coordinate and aggregated over the vector, the improvement transfers beyond pairwise problems to multiclass classification as large as the 1,000-class ImageNet experiment.
  • Classifiers built on linear similarity — KNN with Euclidean or cosine distance, linear SVM, MLP, and decision trees — all show the enhancement, so the effect does not depend on one particular decision rule.

Reading between the lines

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

  • The paper stakes its claim on separated classes and leaves the hard regime unclaimed, which suggests a testable boundary: for genuinely difficult, low-separation problems, coarse quantization should be expected to hurt, and the same discrimination ratio predicts by how much.
  • The analysis is stated per coordinate with a uniform threshold, and the authors note in passing that per-dimension thresholds could raise discrimination further; a dimension-adaptive threshold search is therefore a cheap and direct follow-up experiment.
  • Because the model assumes nearly uncorrelated coordinates, the sharpest stress test is raw, unwhitened features; if the enhancement survives there, the theory is sturdier than its own assumptions, and if not, the real-data successes are explained by the near-decorrelation of the transforms used in the experiments.
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

4 major / 6 minor

Summary. The paper argues that the conventional view equating larger quantization error with worse classification is inadequate, and proposes to evaluate quantized data through feature discrimination, defined as the ratio of expected inter-class squared distance to total intra-class squared distance. The authors model each class as a Gaussian vector whose standardized marginals satisfy X∼N(µ,σ^2) and Y∼N(−µ,σ^2) with µ^2+σ^2=1 (Property 1). They prove (Theorems 4 and 5) that for such scalar variables, binary {0,1} and ternary {0,±1} quantization increase this discrimination whenever the threshold τ satisfies inequalities (8) and (9), respectively, and they show numerically that such thresholds exist for some (µ,σ). Classification experiments on synthetic data and on features from images (YaleB, CIFAR10, ImageNet1000), speech (TIMIT), and text (Newsgroup) are reported, with accuracy curves over threshold sweeps, and the paper concludes that binary and ternary quantization can enhance feature discrimination and classification performance.

Significance. If the vector-level claim were established, the paper would offer a first theoretical account of why aggressive quantization sometimes improves classification, a phenomenon that contradicts the quantization-error heuristic. The scalar theorems are correctly derived: inequalities (8) and (9) are exact algebraic equivalents of D_b>D and D_t>D for the assumed model, with no fitted constants, and the numerical validation reproduces the predicted threshold intervals. The breadth of the real-data evaluation (five datasets, several classifiers) is a further strength. However, the significance as currently stated is limited by the fact that the central applied claim rests on an unproven scalar-to-vector bridge, and the real-data threshold selection is performed on the test set. These issues affect the load-bearing claim of the paper and should be resolved before the result can be considered established for feature vectors.

major comments (4)
  1. [Section 2.2 and Theorems 4-5] The step from scalar variables to feature vectors is asserted but not proved. The sentence in Section 2.2, 'the discrimination between the two random vectors X and Y positively correlates with the discrimination between their each pair of corresponding elements,' is the only bridge between the scalar theorems and the paper's headline claim. For independent dimensions, D_vec = Σ_i W_i D_i / Σ_i W_i, where W_i = E[(X_i - X_i')^2] + E[(Y_i - Y_i')^2]; quantization changes both D_i and W_i, and a dimension whose quantized within-class scatter becomes tiny loses its weight in the aggregate even if its own discrimination improves. Hence D_b^{(i)} > D^{(i)} for every i does not imply the vector-level ratio increases. The synthetic experiments in Section 4.1 report classification accuracy, not the empirical vector-level discrimination ratio, so they do not close this gap. Either a vector-level theorem with explicit conditions, an empirical demonstration that D_vec improves, or a clearly narrowed statement of the claim is required.
  2. [Section 4.2 and Tables 1-2] The real-data thresholds are selected post hoc by sweeping γ and reading test-set accuracy in Figures 5-7, 14-15, and 17-20, so the reported 'enhancement' is determined in hindsight and may not be attainable without test labels. The paper does not specify how the parameters µ and σ needed for Appendix B's gradient-descent solution are estimated from real data, nor whether the thresholds used for the 'Ours' entries in Tables 1-2 are obtained from the training split or from the test set. This weakens the practical claim and makes the real-data validation of the theory incomplete.
  3. [Table 1, CIFAR10 binary row] The binary quantization method 'Ours' achieves 93.54% accuracy versus 94.08% for original data on CIFAR10, contradicting the assertion in Section 4.2.2 that on CIFAR10 both binary and ternary quantization can achieve improved or at least comparable performance. This discrepancy between the table and the figure/caption should be resolved, and the scope of the 'can enhance' claim should be stated precisely.
  4. [Section 2.2 and Figure 16] The real-data validation does not establish the model assumptions. Figure 16 shows marginal histograms of one feature dimension, but the theorems require per-dimension Gaussianity with equal class variances and, for any vector-level extension, independence or a specified dependence structure. The paper does not report how strongly these assumptions are violated on YaleB, TIMIT, Newsgroup, CIFAR10, or ImageNet features, nor whether the enhancing thresholds persist under the measured violations. Without such a check, the generalization from Property 1 to real data remains an assertion.
minor comments (6)
  1. [Introduction] In the phrase 'In the filed of signal processing,' 'filed' should be 'field.'
  2. [References] The Charikar reference contains 'thiry-fourth' instead of 'thirty-fourth.'
  3. [Appendix B.1] In the gradient expression, α = φ(·) and β = φ(·) reuse symbols that were CDFs earlier; these should be written as α′ and β′ or explicitly redefined as densities.
  4. [Figure 15] The label 'Tenary quantization' in panel (h) should be 'Ternary quantization.'
  5. [Equation (18)] The term '2aβ' should be '2αβ.'
  6. [Section 4.2.2] The word 'conductive' in the phrase 'conductive to classification improvement' should be 'conducive.'

Circularity Check

0 steps flagged · score 0.0 of 10

No circular reduction: Theorems 4 and 5 are exact algebraic characterizations, and the real-data experiments are external; the only flagged item is an unproved scalar-to-vector bridge, which is a validity gap rather than circularity.

full rationale

The derivation chain is self-contained. Definition 2 and Definition 3 fix D, Db, and Dt as explicit functions of Gaussian parameters and threshold tau; Theorem 4 and Theorem 5 (with Appendix A) compute closed forms (12), (13), and (18), then solve Db > D and Dt > D into the equivalent inequalities (8) and (9). This is an exact algebraic characterization, not a fitted parameter renamed as a prediction; no empirical constant enters the proof. The numerical validation in Section 3.2 regenerates samples from the same Gaussian model used to derive the formulas, so it verifies arithmetic and sampling consistency rather than providing independent confirmation; that redundancy is not circular because the theorems already stand on the algebra. Real-data experiments in Section 4.2 and Appendix C provide external benchmarks, and the thresholds are searched quantization parameters rather than fitted values that define the discrimination measures. The only self-citation, Lu et al. (2023), is invoked for motivation and for the observation of multiclass improvements, but the current paper's own Figure 20 and real-data results carry that claim, so the citation is not load-bearing. What is genuinely missing is a proof of the Section 2.2 bridge from per-dimension to vector-level discrimination: the statement that vector discrimination 'positively correlates' with element-wise discrimination is an assertion, and the theorems themselves treat scalar dimensions. That is a validity or robustness gap, not a circular reduction, because vector-level discrimination is not defined in terms of element-level discrimination within the derivation. No circular step can be exhibited, so the score is 0.

Assumptions & free parameters 1 free parameters · 4 assumptions · 0 invented entities

The central claim rests on the Gaussian mixture model with symmetric standardization, the element-wise transfer assumption, and Fisher's discrimination rule as a proxy for classification accuracy. No new physical or mathematical entities are introduced. The only practically fitted value is the threshold scan on real data, which is a post-hoc selection rather than a prediction.

free parameters (1)
  • Per-dataset threshold scale γ = Best value found by scanning test accuracy
    In Section 4.2.1 the threshold is set to τ = γ·η and γ is varied across trials; reported improvements are for the optimal scanned value, which is a post-hoc selection on the test set.
assumptions (4)
  • domain assumption Two balanced classes drawn from Gaussian distributions with distinct means and equal variance; after standardization, means are symmetric and µ²+σ²=1 (Property 1).
    This model is the foundation for Theorems 4 and 5; the paper justifies it by citing common use of Gaussian mixtures and the closure of Gaussians under linear operations.
  • domain assumption Vector-level discrimination is positively correlated with per-element discrimination, so analysis can be carried out on a single dimension.
    Stated in Section 2.2 without proof; this transfers the 1D theoretical result to high-dimensional feature vectors.
  • domain assumption Higher Fisher feature discrimination implies easier classification and higher accuracy for KNN and linear SVM.
    The paper adopts Fisher's linear discriminant rule (Section 2.3) but provides no quantitative bound linking D to classification error.
  • domain assumption Real-world features approximately follow per-dimension Gaussian distributions within each class.
    Invoked in Remark 2 and supported by histogram fits in Figure 16; the paper acknowledges real data only approximates this.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Binary and Ternary Quantization Can Enhance Feature Discrimination." pith.science (2026). https://pith.science/paper/EQPWZLFY

@misc{pith2026250413792,
  author       = {Pith},
  title        = {Pith review of: Binary and Ternary Quantization Can Enhance Feature Discrimination},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EQPWZLFY}},
  note         = {Machine review of arXiv:2504.13792}
}
abstract

Quantization is widely applied in machine learning to reduce computational and storage costs for both data and models. Considering that classification tasks are fundamental to the field, it is crucial to investigate how quantization impacts classification performance. Traditional research has focused on quantization errors, assuming that larger errors generally lead to lower classification accuracy. However, this assumption lacks a solid theoretical foundation and often contradicts empirical observations. For example, despite introducing significant errors, $\{0,1\}$-binary and $\{0, \pm1\}$-ternary quantized data have sometimes achieved classification accuracy comparable or even superior to full-precision data. To reasonably explain this phenomenon, a more accurate evaluation of classification performance is required. To achieve this, we propose a direct analysis of the feature discrimination of quantized data, instead of focusing on quantization errors. Our analysis reveals that both binary and ternary quantization can potentially enhance, rather than degrade, the feature discrimination of the original data. This finding is supported by classification experiments conducted on both synthetic and real data.

Figures

Figures reproduced from arXiv: 2504.13792 by the authors.

Figure 1
Figure 1. Consider two classes of data X ∼ N(µ, σ2 ) and Y ∼ N(−µ, σ2 ), with µ = 0.8 and σ 2 = 0.36, as specified in Property 1. The values for the left and right sides of Equations (8) and (9) are provided in (a) and (c) for binary and ternary quantization, respectively; and the discrimination D, Db and Dt statistically estimated with Equations (5), (6) and (7) are illustrated in (b) and (d) for binary and ternary quantizat… view at source ↗
Figure 2
Figure 2. KNN (Euclidean distance) classification accuracy for the 10,000-dimensional [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. KNN (Euclidean distance) classification accuracy for the binary, ternary, and [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (17 more)
Figure 4
Figure 4. Figure 4: KNN (Euclidean distance) classification accuracy, feature discrimination values [PITH_FULL_IMAGE:figures/full_fig_p010_4.png]
Figure 5
Figure 5. Figure 5: Classification accuracy for the binary, ternary, and original data by KNN (Eu [PITH_FULL_IMAGE:figures/full_fig_p012_5.png]
Figure 6
Figure 6. Figure 6: Classification accuracy for the binary, ternary, and original data by KNN (Eu [PITH_FULL_IMAGE:figures/full_fig_p012_6.png]
Figure 7
Figure 7. Figure 7: Classification accuracy for the binary, ternary, and original data by KNN (Eu [PITH_FULL_IMAGE:figures/full_fig_p012_7.png]
Figure 8
Figure 8. Figure 8: Consider the binary quantization on two classes of data [PITH_FULL_IMAGE:figures/full_fig_p022_8.png]
Figure 9
Figure 9. Figure 9: Consider the ternary quantization on two classes of data [PITH_FULL_IMAGE:figures/full_fig_p023_9.png]
Figure 10
Figure 10. Figure 10: KNN (Cosine) classification accuracy for the 10,000-dimensional binary, ternary, [PITH_FULL_IMAGE:figures/full_fig_p024_10.png]
Figure 11
Figure 11. Figure 11: SVM classification accuracy for the 10,000-dimensional binary, ternary, and [PITH_FULL_IMAGE:figures/full_fig_p024_11.png]
Figure 12
Figure 12. Figure 12: KNN (Cosine) classification accuracy for the binary, ternary, and original data [PITH_FULL_IMAGE:figures/full_fig_p025_12.png]
Figure 13
Figure 13. Figure 13: SVM classification accuracy for the binary, ternary, and original data generated [PITH_FULL_IMAGE:figures/full_fig_p025_13.png]
Figure 14
Figure 14. Figure 14: Classification accuracy for the binary, ternary, and original data by KNN (Eu [PITH_FULL_IMAGE:figures/full_fig_p026_14.png]
Figure 15
Figure 15. Figure 15: Classification accuracy for the binary, ternary, and original data by KNN (Cosine [PITH_FULL_IMAGE:figures/full_fig_p027_15.png]
Figure 16
Figure 16. Figure 16: The histogram (blue bar) of the element values across one dimension of the feature [PITH_FULL_IMAGE:figures/full_fig_p028_16.png]
Figure 17
Figure 17. Figure 17: MLP-based binary classification accuracy for the binary, ternary, and original data on three different datasets. The parameter γ corresponds to a quantization threshold τ = γ · η, where η denotes the average magnitude of the feature elements in all feature vectors. Co…
Figure 18
Figure 18. Figure 18: Decision trees-based binary classification accuracy for the binary, ternary, and original data on three different datasets. The parameter γ corresponds to a quantization threshold τ = γ · η, where η denotes the average magnitude of the feature elements in all feature …
Figure 19
Figure 19. Figure 19: Binary classification accuracy for the binary, ternary, and original data in Im￾ageNet1000, using the classifier KNN (Euclidean distance). The parameter γ corresponds to a quantization threshold τ = γ · η, where η denotes the average magnitude of the feature elements …
Figure 20
Figure 20. Figure 20: Multiclass (1000-class) classification accuracy for the binary, ternary, and orig￾inal data in ImageNet1000, using the classifier KNN (Euclidean distance). The parameter γ corresponds to a quantization threshold τ = γ · η, where η denotes the average magnitude of the …

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

41 extracted references · 34 canonical work pages

  1. [1]

    J. S. Baras and S. Dey. Combined compression and classification with learning vector quantization. IEEE Transactions on Information Theory, 45 0 (6): 0 1911--1920, 1999

    work page 1911

  2. [2]

    D. P. Bertsekas. Nonlinear programming. Journal of the Operational Research Society, 48 0 (3): 0 334--334, 1997

    1997

  3. [3]

    M. S. Charikar. Similarity estimation techniques from rounding algorithms. In Proceedings of the thiry-fourth annual ACM symposium on Theory of computing, pages 380--388, 2002

    work page 2002

  4. [4]

    Cortes and V

    C. Cortes and V. Vapnik. Support-vector networks. Machine learning, 20 0 (3): 0 273--297, 1995

    1995

  5. [5]

    Courbariaux, Y

    M. Courbariaux, Y. Bengio, and J.-P. David. Binaryconnect: Training deep neural networks with binary weights during propagations. Advances in neural information processing systems, 28, 2015

    work page 2015

  6. [6]

    Dasgupta, C

    S. Dasgupta, C. F. Stevens, and S. Navlakha. A neural algorithm for a fundamental computing problem. Science, 358 0 (6364): 0 793--796, 2017

    work page 2017

  7. [7]

    J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li, and L. Fei-Fei. ImageNet: A Large-Scale Hierarchical Image Database . In IEEE Conference on Computer Vision and Pattern Recognition, 2009

    work page 2009

  8. [8]

    B. M. Dogahe and M. N. Murthi. Quantization for classification accuracy in high-rate quantizers. In Digital Signal Processing and Signal Processing Education Meeting, pages 277--282. IEEE, 2011

    work page 2011

Show all 41 references
  1. [9]

    R. A. Fisher. The use of multiple measurements in taxonomic problems. Annals of eugenics, 7 0 (2): 0 179--188, 1936

    1936

  2. [10]

    W. M. Fisher, G. R. Doddington, and K. M. Goudie-Marshall. The darpa speech recognition research database: Specifications and status. In Proceedings of DARPA Workshop on Speech Recognition, pages 93--99, 1986

    1986

  3. [11]

    Gholami, S

    A. Gholami, S. Kim, Z. Dong, Z. Yao, M. W. Mahoney, and K. Keutzer. A survey of quantization methods for efficient neural network inference. In Low-Power Computer Vision, pages 291--326. Chapman and Hall/CRC, 2022

    2022

  4. [12]

    R. M. Gray and D. L. Neuhoff. Quantization. IEEE transactions on information theory, 44 0 (6): 0 2325--2383, 1998

    1998

  5. [13]

    Haufe, R

    S. Haufe, R. Tomioka, G. Nolte, K.-R. M \"u ller, and M. Kawanabe. Modeling sparse connectivity between underlying brain sources for EEG/MEG . IEEE transactions on biomedical engineering, 57 0 (8): 0 1954--1963, 2010

    1954

  6. [14]

    K. He, X. Zhang, S. Ren, and J. Sun. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 770--778, 2016

    2016

  7. [15]

    Hoefler, D

    T. Hoefler, D. Alistarh, T. Ben-Nun, N. Dryden, and A. Peste. Sparsity in deep learning: Pruning and growth for efficient inference and training in neural networks. Journal of Machine Learning Research, 22 0 (241): 0 1--124, 2021

    2021

  8. [16]

    Hutchinson, L

    B. Hutchinson, L. Deng, and D. Yu. A deep architecture with bilinear modeling of hidden representations: Applications to phonetic recognition. In IEEE international conference on acoustics, speech and signal processing, pages 4805--4808. IEEE, 2012

    2012

  9. [17]

    Jana and P

    S. Jana and P. Moulin. Optimal design of transform coders and quantizers for image classification. In International Conference on Image Processing, volume 3, pages 841--844. IEEE, 2000

    2000

  10. [18]

    Jana and P

    S. Jana and P. Moulin. Optimal transform coding of gaussian mixtures for joint classification/reconstruction. In Data Compression Conference, pages 313--322. IEEE, 2003

    2003

  11. [19]

    S. Kotz, T. Kozubowski, and K. Podgorski. The Laplace distribution and generalizations: a revisit with applications to communications, economics, engineering, and finance. Springer Science & Business Media, 2012

    2012

  12. [20]

    Krizhevsky and G

    A. Krizhevsky and G. Hinton. Learning multiple layers of features from tiny images. Master's thesis, Department of Computer Science, University of Toronto, 2009

    2009

  13. [21]

    E. Y. Lam and J. W. Goodman. A mathematical analysis of the dct coefficient distributions for images. IEEE transactions on image processing, 9 0 (10): 0 1661--1666, 2000

    2000

  14. [22]

    K. Lang. Newsweeder: Learning to filter netnews. In Machine learning proceedings 1995, pages 331--339. Elsevier, 1995

    1995

  15. [23]

    Larochelle, M

    H. Larochelle, M. Mandel, R. Pascanu, and Y. Bengio. Learning algorithms for the classification restricted boltzmann machine. The Journal of Machine Learning Research, 13 0 (1): 0 643--669, 2012

    2012

  16. [24]

    K. Lee, J. Ho, and D. Kriegman. Acquiring linear subspaces for face recognition under variable lighting. IEEE Transactions on Pattern Analysis and Machine Intelligence, 27 0 (5): 0 684--698, 2005

    2005

  17. [25]

    F. Li, B. Zhang, and B. Liu. Ternary weight networks. arXiv, 1605.04711, 2016

    2016 arXiv

  18. [26]

    D. Lin, S. Talathi, and S. Annapureddy. Fixed point quantization of deep convolutional networks. In International conference on machine learning, pages 2849--2858. PMLR, 2016 a

    2016

  19. [27]

    Z. Lin, M. Courbariaux, R. Memisevic, and Y. Bengio. Neural networks with few multiplications. In International Conference on Learning Representations, 2016 b

    2016

  20. [28]

    W. Lu, M. Chen, K. Guo, and W. Li. Quantization: Is it possible to improve classification? In Data Compression Conference, pages 318--327. IEEE, 2023

    2023

  21. [29]

    Mohamed, T

    A.-r. Mohamed, T. N. Sainath, G. Dahl, B. Ramabhadran, G. E. Hinton, and M. A. Picheny. Deep belief networks using discriminative features for phone recognition. In IEEE international conference on acoustics, speech and signal processing, pages 5060--5063. IEEE, 2011

    2011

  22. [30]

    K. L. Oehler and R. M. Gray. Combining image compression and classification using vector quantization. IEEE transactions on pattern analysis and machine intelligence, 17 0 (5): 0 461--473, 1995

    1995

  23. [31]

    L. E. Peterson. K-nearest neighbor. Scholarpedia, 4 0 (2): 0 1883, 2009

    2009

  24. [32]

    Poor and J

    H. Poor and J. Thomas. Applications of ali-silvey distance measures in the design generalized quantizers for binary decision systems. IEEE Transactions on Communications, 25 0 (9): 0 893--900, 1977

    1977

  25. [33]

    H. Qin, R. Gong, X. Liu, X. Bai, J. Song, and N. Sebe. Binary neural networks: A survey. Pattern Recognition, 105: 0 107281, 2020

    2020

  26. [34]

    J. R. Quinlan. Induction of decision trees. Machine learning, 1: 0 81--106, 1986

    1986

  27. [35]

    D. E. Rumelhart, G. E. Hinton, and R. J. Williams. Learning representations by back-propagating errors. nature, 323 0 (6088): 0 533--536, 1986

    1986

  28. [36]

    Simonyan and A

    K. Simonyan and A. Zisserman. Very deep convolutional networks for large-scale image recognition. arXiv preprint arXiv:1409.1556, 2014

    2014 arXiv

  29. [37]

    Srinivasamurthy and A

    N. Srinivasamurthy and A. Ortega. Reduced complexity quantization under classification constraints. In Data Compression Conference, pages 402--411. IEEE, 2002

    2002

  30. [38]

    Torralba and A

    A. Torralba and A. Oliva. Statistics of natural image categories. Network: computation in neural systems, 14 0 (3): 0 391--412, 2003

    2003

  31. [39]

    M. J. Wainwright and E. P. Simoncelli. Scale mixtures of gaussians and the statistics of natural images. In Proceedings of the 12th International Conference on Neural Information Processing Systems, 1999

    1999

  32. [40]

    H. Wang, S. Ma, L. Dong, S. Huang, H. Wang, L. Ma, F. Yang, R. Wang, Y. Wu, and F. Wei. Bitnet: Scaling 1-bit transformers for large language models. arXiv preprint arXiv:2310.11453, 2023

    2023 arXiv

  33. [41]

    Weiss and W

    Y. Weiss and W. T. Freeman. What makes a good model of natural images? In IEEE Conference on Computer Vision and Pattern Recognition, pages 1--8. IEEE, 2007

    2007

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.