Pith. sign in

REVIEW 3 major objections 6 minor 23 references

Multi-hypothesis classifier

T0 review · 3 major / 6 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read The paper claims that a confidence-weighted combination of three nearest-neighbour classifiers raises Tibetan character recognition accuracy from 84.06% to 88.63%.

desk verdict The headline accuracy gains vanish once the discarded samples are counted as errors; the Tibetan OCR setting is real, but the central comparison is not valid. read the letter →

arxiv 1908.07857 v1 pith:JNZ6R44A submitted 2019-08-20 cs.LG

Sayantan Sengupta , Sudip Sanyal This is my paper · ORCID
classification cs.LG
keywords multi-hypothesisclassifierensembleconfidencematrixnearestneighbourGaborfilterprincipalcomponentanalysisTibetancharacterrecognitionweightedmajorityvoting
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 thesis tries to establish that a small committee of three deliberately different nearest-neighbour classifiers, each built on a different representation of Tibetan character images, can be fused by confidence values so that the committee beats every one of its members. The three base classifiers use raw pixels, centroid-distance shape features, and Gabor-filter features reduced by PCA, and the paper devises five fusion hypotheses based on confidence matrices. It reports that all five combinations improve on the best individual classifier, with the top result of 88.63% against an 84.06% single-classifier baseline. The broader point is that combining strong classifiers is worthwhile, not only combining many weak ones.

What carries the argument

The central object is the confidence matrix: for each base classifier, each diagonal entry of its confusion matrix is divided by the column sum, giving a per-class belief that the classifier is right when it predicts that class. Fusion works by treating this matrix as a lookup table, adding or comparing confidence values for the class labels proposed by the three classifiers, and then picking the label with the highest weighted support. Hypothesis #2 substitutes a sample-level confidence: in nearest-neighbour search, the ratio of the second-best distance to the best distance measures how decisive the top match is, and this sample-level confidence drives the vote. This machinery is what converts three separate label predictions into a single ensemble decision.

What would settle it

Recompute the ensemble accuracy for Hypothesis #2 on the full validation set without discarding samples where all three classifiers err, and compute the three individual classifiers' accuracies on that same validation set. If the unconditional ensemble accuracy no longer exceeds 84.06%, or the gain shrinks once baseline and ensemble are measured on identical samples, the reported improvement is an artifact of conditioning. The 64 discarded sample numbers listed in the paper provide the exact subset to restore.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central claim is that a multi-hypothesis classifier improves accuracy in all five tested fusion strategies. Each base classifier is treated as a hypothesis whose confusion matrix can be converted into a confidence matrix; the ensemble then consults these confidence values and combines the individual predictions by weighted majority or maximum confidence. The best variant, Hypothesis #2, builds confidence from the ratio between the distance to the best match and the distance to the second-best match in nearest-neighbour search, and reaches 88.63%, compared with 84.06% for the strongest single classifier. The paper also observes that certain visually similar Tibetan character classes are consistently confusable and remain errors even after fusion.

Load-bearing premise

The load-bearing assumption is that every counted validation sample is correctly labelled by at least one of the three classifiers; the paper discards all samples where all three are wrong and reports accuracy only on the remaining samples, so the headline 88.63% is a conditional success rate rather than an accuracy over all validation samples.

Editorial extensions

If this is right

  • A committee of strong, feature-diverse classifiers can outperform each of its members on the same data.
  • The decisive-match distance ratio in nearest-neighbour search is a usable confidence signal for voting, not just a ranking criterion.
  • Fusion cannot repair confusions that all base classifiers share, such as visually near-identical Tibetan character classes.
  • The reported gain is measured only on validation samples where at least one base classifier is correct, so the ensemble's operational accuracy on every sample is a separate question.
  • Adding more diverse classifiers or replacing random sample selection with deliberate sampling could further raise ensemble accuracy, as the paper's future-work discussion suggests.

Reading between the lines

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

  • A fair test of the claimed gain would compute both the ensemble and the 84.06% baseline on the identical, complete validation set, including samples every classifier gets wrong; until then the reported gain is conditional, not guaranteed.
  • The confidence-matrix design is generic, so the same five hypotheses could be run on printed Latin, Devanagari, or Chinese character datasets with fixed train/test splits; if the conditional gain disappears there, the method is specific to this Tibetan dataset.
  • The failure cases named by the paper suggest a testable extension: add a rejection stage that sends low-confidence samples to a dedicated pairwise discriminator rather than forcing the ensemble to choose among visually similar classes.
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

3 major / 6 minor

Summary. The manuscript tests five ensemble strategies for combining three nearest-neighbour classifiers on a 64-class Tibetan character recognition task. The classifiers use raw pixels, centroid-distance structural features, and Gabor-filter responses reduced by PCA; the five 'hypotheses' differ in how per-class confidence weights are derived and combined. The paper's central claim, in Table 5.1, is that every hypothesis improves on the best single classifier, with Hypothesis #2 reaching 88.63% versus 84.06% for the best base classifier. The thesis also documents the data distribution, confusion matrices, and lists of discarded validation samples.

Significance. If the reported comparison were valid, the work would give a simple demonstration that confidence-weighted voting of three nearest-neighbour classifiers improves character-recognition accuracy, and the explicit recording of discarded samples would be a useful auditing feature. However, the central comparison is not valid: ensemble accuracies are computed on a filtered validation subset while the baseline accuracies are unconditional and from a different split. The paper also selects hyperparameters on the test set and never tests its stated independence assumption. The manuscript does not, as presented, establish the claimed improvement, and the experimental design cannot be repaired without changing the conclusion.

major comments (3)
  1. [Sections 4.1.8, 4.2.2, 4.3.6, 4.4.6, 4.5.1; Table 5.1] The headline accuracies are computed on the subset of validation samples for which at least one base classifier is correct, not on the full validation set. For Hypothesis #1, Section 4.1.8 reports 223 correct out of 255 such samples (87.45%), and the 65 samples for which all three classifiers were wrong are explicitly listed and discarded; Sections 4.2.2, 4.3.6, 4.4.6, and 4.5.1 do the same. The baseline accuracies in Table 5.1 (e.g., 84.06% for Classifier #1) are unconditional accuracies on the separate test split. Counting the discarded samples as errors gives 223/320=69.69% for Hypothesis #1 and 226/320=70.63% for Hypothesis #2, both below the 84.06% baseline; the corresponding unconditional values for Hypotheses #3-#5 are 194/320=60.63%, 190/320=59.38%, and 191/320=59.69%, all below the 77.81% baseline used for those hypotheses. The conclusion in Chapter 5 that 'the overall accuracy increases' is therefore unsupported. A valid comparison requires either counting discarded samples as errors on the full validation set or reporting the base-classifier accuracies on the same filtered subset with the coverage rate.
  2. [Sections 3.3.3, 4.1.3, 4.1.4, 4.1.6] The same test split is used for multiple purposes: Tables 4.1 and 4.2 select the PCA component count (110) and Gabor kernel size (7) by maximizing accuracy, Section 4.1.6 builds the confidence matrices from the classifiers' confusion matrices on that split, and Table 5.1 reports the base-classifier accuracies from that split. The ensemble is then evaluated on the validation split. This protocol provides no independent estimate of generalization: the reported base accuracies are optimistic because the hyperparameters were tuned on the test set, and the confidence weights are not validated on data unseen during their construction. A separate validation set should be used for model selection and confidence estimation, or nested cross-validation should be employed.
  3. [Chapter 3 (basic assumption) and Chapter 6] The manuscript states as a basic assumption that the classifiers are conditionally independent and do not make the same mistakes, but it provides no diversity measure or error-correlation analysis, and Chapter 6 admits that this property was neither achieved nor proven. The coverage numbers themselves (255/320 for Hypotheses #1-#2, 233/320 for Hypothesis #3, 227/320 for Hypothesis #4, 228/320 for Hypothesis #5) show that a substantial fraction of validation samples are missed by all three classifiers, which is direct evidence that errors are not complementary in the strong sense required by the motivation. The independence assumption needs to be tested (e.g., with pairwise Q-statistics or disagreement measures) before the success of the method can be attributed to diversity.
minor comments (6)
  1. [Section 4.2.2 and Table 5.1] Hypothesis #2 accuracy is given as 88.62% in the text and 88.63% in the table; please reconcile.
  2. [Sections 4.3.3, 4.4.3 and Table 5.1] Classifier #3 accuracy for Hypothesis #3 is 60.62% in Section 4.3.3 and 60.63% in Table 5.1; for Hypothesis #4 it is 4.37% in Section 4.4.3 and 4.38% in Table 5.1; please reconcile these values.
  3. [Sections 3.3.3 and 4.1.3] Section 3.3.3 says PCA reduces the Gabor feature dimension to 100 as the optimum, while Section 4.1.3 concludes that 110 components is optimal; the text should be consistent about the actual setting used for Classifier #3.
  4. [References] The reference list is incomplete: the citation for Freund and Schapire/Quinlan in Section 2.2 is [5], but the list jumps from [4] to [6] without an entry for [5].
  5. [Section 3.1] The 15/5/5 train/test/validation split is described only at the level of counts; please specify how the random selection was performed and whether the split was stratified, to make the experiments reproducible.
  6. [General] There are numerous typographical and OCR artifacts (e.g., 'Bressenham' for Bresenham, broken equations in Chapter 2) that should be cleaned up in a revised version.

Circularity Check

1 steps flagged · score 6.0 of 10

Reported ensemble gains are computed on the subset where at least one base classifier is correct, making the comparison to unconditional baselines an artifact of the metric definition.

  1. self definitional [Section 4.1.8 (repeated in 4.2.2, 4.3.6, 4.4.6, 4.5.1); Table 5.1]
    "Now, for our hypothesis to work decently at least one classifier should give the correct prediction. The number of such samples from the above observation is 255. Among them, the number of correctly classified samples by the hypothesis is 223. So our final accuracy stands at 87.45 %."

    The reported 'final accuracy' is defined conditionally on the base classifiers' own success: the denominator is the number of validation samples for which at least one of the three base classifiers is correct, and samples where all three are wrong are discarded. The baseline accuracies in Table 5.1 (e.g., 84.06% for Classifier #1) are unconditional accuracies on the test split. Thus the claimed 'overall accuracy increases' compares a conditional probability, P(ensemble correct | at least one base classifier correct), with unconditional accuracies, so the gain is built into the metric rather than demonstrated by the ensemble.

full rationale

The ensemble construction itself is largely self-contained: confidence matrices are built from each classifier's confusion matrix on the test split and then applied to a separate validation split, with no self-citation chain or imported uniqueness theorem carrying the argument. The central circularity is in the evaluation metric. The paper's headline conclusion, 'In all the five cases above, the overall accuracy increases,' rests on accuracies whose denominators are counts of validation samples on which at least one base classifier is already correct. Because the discarded samples are exactly those on which all base classifiers fail, the ensemble's conditional accuracy is inflated by construction relative to the unconditional baselines used in Table 5.1. This is not a case of fitted parameters being renamed as predictions, nor of an ansatz smuggled in by citation; it is a self-referential performance measure that makes the main quantitative claim an artifact of how accuracy is defined. Apart from this evaluation artifact, the thesis does not exhibit other circular steps.

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

The central claim rests on hyperparameters tuned to the test set, an unverified independence assumption, and a filtered evaluation subset. No new physical or mathematical entities are introduced.

free parameters (4)
  • Number of PCA components retained = 110
    Chosen by scanning 20 to 140 components and selecting the value with the highest classifier #3 test accuracy (Section 4.1.3).
  • Gabor kernel size = 7
    Chosen by scanning kernel sizes 3 to 10 and selecting the value with the highest classifier #3 test accuracy (Section 4.1.4).
  • Gabor filter parameters (frequency, sigma_x, sigma_y) = not specified
    The Gabor bank uses four orientations (0, 45, 90, 135) but the frequency and standard deviations are not reported, leaving implicit free parameters.
  • Sample counts per class = 25 (15 train, 5 test, 5 validation)
    Chosen by hand to impose class balance; the split sizes affect all reported accuracies.
assumptions (4)
  • ad hoc to paper The three classifiers are conditionally independent and do not make the same mistakes.
    Stated in Section 1.1.2 and repeated in Chapter 3 as 'the basic assumption still holds true... there is no guarantee to it.' It is load-bearing for the expected ensemble gain.
  • domain assumption Test-set confusion matrices reliably estimate classifier confidence on validation samples.
    All five hypotheses compute confidence matrices from the test-set confusion matrices and apply them to validation predictions (Sections 3.4-3.8). This assumes no distribution shift between test and validation sets.
  • domain assumption Random sample selection produces representative training, test, and validation splits.
    Section 3.1 uses a random function to select samples; the authors later note in Future Scope that sampling theory would have been better, so representativeness is assumed without evidence.
  • domain assumption Euclidean distance on binarized 32x32 pixel images is a suitable similarity measure for Tibetan characters.
    All three classifiers use nearest-neighbor with Euclidean distance (Section 3.2); the suitability of this geometry for Tibetan character shapes is not justified.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Multi-hypothesis classifier." pith.science (2026). https://pith.science/paper/JNZ6R44A

@misc{pith2026190807857,
  author       = {Pith},
  title        = {Pith review of: Multi-hypothesis classifier},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JNZ6R44A}},
  note         = {Machine review of arXiv:1908.07857}
}
read the original abstract

Accuracy is the most important parameter among few others which defines the effectiveness of a machine learning algorithm. Higher accuracy is always desirable. Now, there is a vast number of well established learning algorithms already present in the scientific domain. Each one of them has its own merits and demerits. Merits and demerits are evaluated in terms of accuracy, speed of convergence, complexity of the algorithm, generalization property, and robustness among many others. Also the learning algorithms are data-distribution dependent. Each learning algorithm is suitable for a particular distribution of data. Unfortunately, no dominant classifier exists for all the data distribution, and the data distribution task at hand is usually unknown. Not one classifier can be discriminative well enough if the number of classes are huge. So the underlying problem is that a single classifier is not enough to classify the whole sample space correctly. This thesis is about exploring the different techniques of combining the classifiers so as to obtain the optimal accuracy. Three classifiers are implemented namely plain old nearest neighbor on raw pixels, a structural feature extracted neighbor and Gabor feature extracted nearest neighbor. Five different combination strategies are devised and tested on Tibetan character images and analyzed

Figures

Figures reproduced from arXiv: 1908.07857 by the authors.

Figure 1
Figure 1. Serial Architecture...................................................................... [PITH_FULL_IMAGE:figures/full_fig_p011_1.png] view at source ↗
Figure 2
Figure 2. Flowchart of Bagging.......................................................... [PITH_FULL_IMAGE:figures/full_fig_p011_2.png] view at source ↗
Figure 1.1
Figure 1.1. Several classifiers in different feature spaces. [PITH_FULL_IMAGE:figures/full_fig_p015_1_1.png] view at source ↗
Figures from the paper (35 more)
Figure 1
Figure 1. Figure 1 [PITH_FULL_IMAGE:figures/full_fig_p016_1.png]
Figure 1.3
Figure 1.3. Figure 1.3: D* is the best performing individual classifier; the space of all [PITH_FULL_IMAGE:figures/full_fig_p017_1_3.png]
Figure 1.4
Figure 1.4. Figure 1.4: D* is the best classifier for the data; the space of all classifiers is shown by the outer curve; the dashed lines represents the hypothetical path or trajectories for the classifiers during training [PITH_FULL_IMAGE:figures/full_fig_p017_1_4.png]
Figure 1.5
Figure 1.5. Figure 1.5: Banana dataset; Optimal classifier is nonlinear. [PITH_FULL_IMAGE:figures/full_fig_p018_1_5.png]
Figure 1.6
Figure 1.6. Figure 1.6: Serial architecture. EXPERT 1 EXPERT 2 ................. EXPERT N [PITH_FULL_IMAGE:figures/full_fig_p019_1_6.png]
Figure 1.7
Figure 1.7. Figure 1.7: Parallel architecture. 1.4.1.3 Hybrid [PITH_FULL_IMAGE:figures/full_fig_p020_1_7.png]
Figure 1
Figure 1. Figure 1 [PITH_FULL_IMAGE:figures/full_fig_p020_1.png]
Figure 1.9
Figure 1.9. Figure 1.9: Single feature set, different classifiers. [PITH_FULL_IMAGE:figures/full_fig_p021_1_9.png]
Figure 1.10
Figure 1.10. Figure 1.10: Different feature space. 1.4.2 Fixed Combination Strategy Some of the popular fixed combination rules are Product Rule (minimum), Sum rule (mean), Median rule, Majority vote and Maximum rule. These rules will be discussed in the later chapter in details. Some featur…
Figure 1.11
Figure 1.11. Figure 1.11: Trained combiners Trained combiners perform potentially better than the fixed rules because of their flexibility. Trained rules are claimed to be more suitable than the fixed ones for classifiers correlated or exhibiting different performances. Trained combiner negl…
Figure 2.1
Figure 2.1. Figure 2.1: Flowchart of Boosting. Now, having established the analogy lets dive in to the technical know-how of this method. The main idea is to combine many weak classifiers to produce powerful committee. So the classifiers are produced sequentially, one after the other. Each …
Figure 2.2
Figure 2.2. Figure 2.2: Example of data sampling in subsequent rounds of boosting. [PITH_FULL_IMAGE:figures/full_fig_p026_2_2.png]
Figure 2.3
Figure 2.3. Figure 2.3: Flow chart of Bagging The performance of the bagging depends on the stability of the base classifier. If a base classifier is unstable, bagging helps to reduce the errors associated with random fluctuations in the training data. If the base classifier is stable, bagg…
Figure 3.1
Figure 3.1. Figure 3.1: Calculating the distance of the last black pixel along the vector drawn from the [PITH_FULL_IMAGE:figures/full_fig_p054_3_1.png]
Figure 3.2
Figure 3.2. Figure 3.2: A visual representation of a 2-D Gabor filter [PITH_FULL_IMAGE:figures/full_fig_p056_3_2.png]
Figure 3.3
Figure 3.3. Figure 3.3: Gabor Filter bank with four different orientations (0, 45, 90, 135) are shown [PITH_FULL_IMAGE:figures/full_fig_p057_3_3.png]
Figure 3.4
Figure 3.4. Figure 3.4: Regression (left) v/s PCA (right). The Figure in the left shows regression whose objective is to minimise the output error. On the other hand PCA minimises the projection error. The basic steps to implement PCA are as follows: 1) Get some data 2) Subtract the mean 3)…
Figure 4.1
Figure 4.1. Figure 4.1: Confusion matrix of classifier-1(Nearest neighbour algorithm) [PITH_FULL_IMAGE:figures/full_fig_p062_4_1.png]
Figure 4.2
Figure 4.2. Figure 4.2: Confusion matrix of classifier-2 (Centroid feature extraction method) [PITH_FULL_IMAGE:figures/full_fig_p063_4_2.png]
Figure 4.3
Figure 4.3. Figure 4.3: Confusion matrix of classifier #3 Conclusion: The accuracy obtained is 74.37 % [PITH_FULL_IMAGE:figures/full_fig_p066_4_3.png]
Figure 4.4
Figure 4.4. Figure 4.4: Confidence matrix of the 3 classifiers [PITH_FULL_IMAGE:figures/full_fig_p067_4_4.png]
Figure 4.5
Figure 4.5. Figure 4.5: Summary of the final hypothesis [PITH_FULL_IMAGE:figures/full_fig_p070_4_5.png]
Figure 4.6
Figure 4.6. Figure 4.6: Confidence of each sample for different classifiers. [PITH_FULL_IMAGE:figures/full_fig_p075_4_6.png]
Figure 4.7
Figure 4.7. Figure 4.7: Summary of Hypothesis #2 [PITH_FULL_IMAGE:figures/full_fig_p078_4_7.png]
Figure 4.8
Figure 4.8. Figure 4.8: Confusion Matrix of Classifier #1 Conclusion: Accuracy of Classifier #1 is 77.81 % [PITH_FULL_IMAGE:figures/full_fig_p081_4_8.png]
Figure 4.9
Figure 4.9. Figure 4.9: Confusion matrix of classifier #2 Conclusion: Accuracy of Classifier #2 = 77.18 % [PITH_FULL_IMAGE:figures/full_fig_p082_4_9.png]
Figure 4.10
Figure 4.10. Figure 4.10: Confusion matrix of classifier #3 Conclusion: Accuracy of Classifier #3 = 60.62 % [PITH_FULL_IMAGE:figures/full_fig_p083_4_10.png]
Figure 4.11
Figure 4.11. Figure 4.11: Confidence matrix of hypothesis #3 [PITH_FULL_IMAGE:figures/full_fig_p084_4_11.png]
Figure 4.12
Figure 4.12. Figure 4.12: Summary of Hypothesis #3 [PITH_FULL_IMAGE:figures/full_fig_p087_4_12.png]
Figure 4.13
Figure 4.13. Figure 4.13: Confusion matrix of classifier #1. Conclusion: Accuracy of classifier #1 = 77.81 % [PITH_FULL_IMAGE:figures/full_fig_p090_4_13.png]
Figure 4.14
Figure 4.14. Figure 4.14: Confusion matrix of classifier #2. Conclusion: Accuracy of classifier #2 = 63.12 % [PITH_FULL_IMAGE:figures/full_fig_p091_4_14.png]
Figure 4.15
Figure 4.15. Figure 4.15: Confusion matrix of classifier #3 Conclusion: Accuracy of classifier #3 = 4.37 %. Most of the samples are getting incorrectly classified [PITH_FULL_IMAGE:figures/full_fig_p092_4_15.png]
Figure 4.16
Figure 4.16. Figure 4.16: Confidence matrix of Hypothesis #4 [PITH_FULL_IMAGE:figures/full_fig_p093_4_16.png]
Figure 4.17
Figure 4.17. Figure 4.17: Summary of Hypothesis #4 [PITH_FULL_IMAGE:figures/full_fig_p096_4_17.png]
Figure 4.18
Figure 4.18. Figure 4.18: Summary of Hypothesis #5 [PITH_FULL_IMAGE:figures/full_fig_p101_4_18.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

23 extracted references · 23 canonical work pages

  1. [1]

    In the above equation  denotes the logical AND operator or binary multiplication and in the following equations  denotes the logical OR operator or binary summation

    (  MxE , Otherwise The above equation means that if all K classifiers decide that sample x belongs to class C j unanimously, then classifier E decides that x comes from C j , otherwise it rejects x. In the above equation  denotes the logical AND operator or binary multiplication and in the following equations  denotes the logical OR operator or binary...

    work page

  2. [2]

    In other words, the classifiers that reject the sample x will have no impact on the combined classifier unless all the classifier reject x

    (  MxE , Otherwise Multi-Hypothesis Classifier 2014 22 Indian Institute of Information Technology, Allahabad The above modified equation, means that sample x belongs to a particular class C k as long as some of the classifiers support that class C k and no other classifier support a different class. In other words, the classifiers that reject the sample...

    work page 2014

  3. [3]

    drugpa kunleg

    (  MxE , Otherwise Where 0 < α <= 1. As the number of classifiers (K) are constant. The vote of the second maximum is taken as the implicit objections to the label j. 2.6 Combination of Classifiers in Dempster-Shafer Formalism This combination technique is useful only when the output information provided by each classifier is in abstract form, i.e. only...

    work page 1998

  4. [4]

    Check for the book with the minimum sample

    work page

  5. [5]

    If ( Number of samples < 6) ; Take all the samples from that book

    work page

  6. [6]

    Else ; Take only 6 samples from that book

    work page

  7. [7]

    IF ( Last book? ), GOTO Step 6

    work page

  8. [8]

    Go to the book with next minimum sample and repeat steps 2 &3

    work page

Show all 23 references
  1. [9]

    Which samples to collect from each class of each book, given the number of samples to be collected from each book for each class?

    Sum all the samples taken so far (sum) and take 25 - sum number of samples from the last book. Thus we have got a somewhat equally distributed samples from each book. Now, the next dilemma is - "Which samples to collect from each class of each book, given the number of samples...

    2014

  2. [10]

    Calculate the Covariance matrix

  3. [11]

    Calculate the Eigen values and corresponding Eigen vectors of the covariance matrix

  4. [12]

    The last step is the deciding step in maintaining the required variance of the data

    Choosing the required number of components and forming the feature vector. The last step is the deciding step in maintaining the required variance of the data. After computing the Eigen values of the covariance matrix, it is found that only few Eigen values are significant com...

    2014

  5. [15]

    Danian Zheng, Ya ŶŶaŶ )hao, JiadžiŶ WaŶg,͟ Featuƌes edžtƌaĐtioŶ usiŶg a Gaďoƌ filteƌ faŵilLJ,͟ Proceedings of 6 th IASTED International Conference, Signal and Image processing, August 23- 25,2004, Honolulu, Hawai, USA Multi-Hypothesis Classifier 2014 95 Indian Institute of Inform...

    2004

  6. [16]

    Increase number of Classifiers: Some more number of classifiers can be used such as SVM, ANN, Fringe maps and the overall effect of using theses can be analysed

  7. [17]

    Sampling techniques: One particular observation drawn from this research is the conflicts of results given on the test set and the validation set by a particular classifier for their respective sample numbers, i.e. for samples belonging to a particular class label, the predict...

  8. [18]

    statistically independent

    Forming the perfect ensemble: The basic assumption made on this research is that the classifiers are mutually exclusive i.e. statistically independent. Although we haven’t exactly explored how to achieve t his or how to prove if they are mutually complementary. Work can be pur...

    2014

  9. [19]

    L. I . Kuncheva, Combining pattern Classifiers: methods and algorithms. New Jersey: J ohn Wiley and Sons, 2004. [ϯ]K. TuƌŶeƌ aŶd J. Ghosh, ͞Eƌƌoƌ CoƌƌelatioŶ aŶd eƌƌoƌ ƌeduĐtioŶ iŶ eŶseŵďle Đlassifieƌs,͟ Connection Sciences, vol. 8, no. ¾, pp. 385-404, 1996. [ϰ] L. BƌeiŵaŶ, ͞B...

    2004

  10. [20]

    Freund and R.E

    Y. Freund and R.E. Scha piƌe, ͞ A DeĐisioŶ -Theoretic Generalization of on-Line Learning and an AppliĐatioŶ to BoostiŶg,͟ J. Coŵputeƌ aŶd “LJsteŵ “ĐieŶĐes, ǀol. ϱϱ, Ŷo. ϭ, pp. ϭϭϵ-139, 1997. [ϳ] M. “kuƌiĐhiŶa, L. I. KuŶĐheǀa, aŶd R. P. DuiŶ, ͞BaggiŶg aŶd BoostiŶg foƌ the Ŷeaƌes...

    1997

  11. [21]

    Kittler, M

    J. Kittler, M. Hatef, R. Duin and J. Matas, ͞OŶ CoŵďiŶiŶg Classifieƌs,͟ IEEE TƌaŶsaĐtioŶs oŶ PatteƌŶ Analysis and Machine Intelligence, 20(3), March 1998, pp. 226-239. [ϵ] Lei Xu, Adaŵ KƌLJzak, ChiŶg Y “ueŶ, ͞Methods of ĐoŵďiŶiŶg Đlassifieƌs aŶd theiƌ appliĐatioŶs to HaŶdǁƌitiŶ...

    1998

  12. [23]

    http://en.wikipedia.org/wiki/Principal_component_analysis. [ϭϳ] Kaƌthika VeŶkatƌaŵaŶi,͟ Optiŵal Đlassifieƌ EŶseŵďles foƌ iŵpƌoǀed BioŵetƌiĐ ǀeƌifiĐatio Ŷ,͟ PhD dissertation, department of Electrical and Computer engineering, Carnegie Mellon University

  13. [135]

    average distances for each class

    which will generate a feature vector of size 4096. After this, we apply PCA to reduce the dimension from 4096 to 100 (optimum). After the feature vectors are formed, this is fed to our plain old nearest neighbour algorithm and the corresponding confusion matrix is created. Now...

    2014

  14. [194]

    Thus the final accuracy is 83.26%. ACCURACY = 83.26 % The samples discarded are: 18, 44, 50, 51, 52, 53, 54, 58, 59, 65, 66, 67, 68, 69, 74, 76, 87, 119, 124, 136, 141, 142, 143, 144, 158, 159, 170, 171, 172, 173, 182, 185, 186, 187, 192, 193, 195, 197, 199, 204, 215, 216, 217...

    2014

  15. [255]

    So the accuracy comes to 88.62 %

    And the number of correctly predicted samples by the hypothesis is 226. So the accuracy comes to 88.62 %. ACCURACY = 88.62 % The samples discarded are: 44, 54, 59, 65, 66, 67, 68, 69, 74, 76, 87, 90, 91, 92, 93, 94, 136, 158, 159, 170, 171, 172, 185, 186, 187, 193, 195, 204, 2...

    2014

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.