Classifiers Based on Inverted Distances

Jirina, Marcel

doi:10.5772/13971

Cited by 10 publications

(6 citation statements)

References 19 publications

(17 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…But because our met h o d (Eb kNN) i s not related in any way to either of these previous works, I merely make a mention of them here and do not discuss t h em. Also because our work (EbkNN) is neither an advancement over these works [27,2,28,29,30,31,32,33] n o r is d eriv ed from them, we do not immediately conclude or claim in this paper that EbkNN outperforms a subset of these and is inferior to the rest. Moreover, we do not state in this paper any mathematical theorem giving the lower and upper bounds on the probability of error of EbkNN.…”

Section: Introductionmentioning

confidence: 83%

“…They took k values so large as 30, 45 and 65; and fou n d t hat the performance of the kNN on Reuters versions 3 and 4 was one of the best. There are a plethora of other literature t hat solves the problem of optimizing performance of kNN by fixing k [30,31,32,33]. But because our met h o d (Eb kNN) i s not related in any way to either of these previous works, I merely make a mention of them here and do not discuss t h em.…”

Section: Introductionmentioning

confidence: 96%

See 1 more Smart Citation

Entropy Based k Nearest Neighbor Pattern Classification (EbkNN): En-route to Achieving a High Accuracy in Breast Cancer Diagnosis

Sinha

Sinha²

2020

AJAS

View full text Add to dashboard Cite

Entropy based k-Nearest Neighbor pattern classification (EbkNN) is a variation of the conventional k-Nearest Neighbor rule of pattern classification, which exclusively optimizes the value of k-neighbors for each test data based on the calculations of entropy. The formula for entropy used in EbkNN is the one that has been defined popularly in information theory for a set of n different types of information (class) attached to a total of m objects (data points) with each object defined by f features. In EbkNN that value of k is chosen for discrimination of given test data for which the entropy is the least non-zero value. Other rules of conventional kNN are retained in EbkNN. It is concluded that EbkNN works best for binary classification. It is computationally prohibitive to use EbkNN for discriminating the data points of the test dataset into number of classes greater than two. The biggest advantage of EbkNN vis-à-vis the conventional kNN is that in one single run of EbkNN algorithm we get optimum classification of test data. But conventional kNN algorithm has to be run separately for each of the selected range of values of k, and then the optimum k to be chosen from amongst them. We also tested our EbkNN method on WDBC (Wisconsin Diagnostic Breast Cancer) dataset. There are 569 instances in this dataset and we made a random choice of first 290 instances as training dataset and the rest 279 instances as test dataset. We got an exceptionally remarkable result with EbkNN method- accuracy close to 100% and better than the ones got by most of the other researchers who worked on WDBC dataset.

show abstract

Section: Introductionmentioning

confidence: 83%

Section: Introductionmentioning

confidence: 96%

Entropy Based k Nearest Neighbor Pattern Classification (EbkNN): En-route to Achieving a High Accuracy in Breast Cancer Diagnosis

Sinha

Sinha²

2020

AJAS

View full text Add to dashboard Cite

show abstract

“…For the classification of the pure isomers (Figure a), 300 data points, i.e., 100 data points of each isomer, were used. The K -value in kNN is set to 40, corresponding to the square root of the number of points, here 1600, following a general recommendation for setting K . − (Varying the K -value by 50% did not result in a significant change of the classification results.) The kNN classification was performed using 10-fold cross-validation, with 90% of the data points used as the training set (i.e., 1440 points) and 10% as the test set (160 points).…”

Section: Methodsmentioning

confidence: 99%

VOC Mixture Sensing with a MOF Film Sensor Array: Detection and Discrimination of Xylene Isomers and Their Ternary Blends

et al. 2022

View full text Add to dashboard Cite

Detection and recognition of volatile organic compounds (VOCs) are crucial in many applications. While pure VOCs can be detected by various sensors, the discrimination of VOCs in mixtures, especially of similar molecules, is hindered by crosssensitivities. Isomer identification in mixtures is even harder. Metal−organic frameworks (MOFs) with their well-defined, nanoporous, and versatile structures have the potential to improve the VOC sensing performance by tailoring the adsorption affinities. Here, we detect and identify ternary xylene isomer mixtures by using an array of six gravimetric, quartz crystal microbalance (QCM)-based sensors coated with selected MOF films with different isomer affinities. We use classical molecular simulations to provide insights into the sensing mechanism. In addition to the attractive interaction between the analytes and the MOF film, the isomer discrimination is caused by the rigid crystalline framework sterically controlling the access of the isomers to different adsorption sites in the MOFs. The sensor array has a very low limit of detection of 1 ppm for each pure isomer and allows the isomer discrimination in mixtures. At 100 ppm, 16 different ternary o−p−m-xylene mixtures were identified with high classification accuracy (96.5%). This work shows the unprecedented performance of MOF-sensor arrays, also referred to as MOF-electronic nose (MOF-enose), for sensing VOC mixtures. Based on the study, guidelines for detecting and discriminating complex mixtures of volatile molecules are also provided.

show abstract

“…We retrieve the latent space vectors for all flipped training images as well and used only a single image per scene (i.e., not all 10 variations). We chose k = √ N = 115, where N is the size of the training data together with its flipped version [84]. We froze the same layers of the pretrained models for fine-tuning the later layers in case of classification models or to train our autoencoder using it as an extractor.…”

Section: Sviro To Ticammentioning

confidence: 99%

Autoencoder and Partially Impossible Reconstruction Losses

Cruz

Taetz

Stifter³

et al. 2022

Sensors

View full text Add to dashboard Cite

The generally unsupervised nature of autoencoder models implies that the main training metric is formulated as the error between input images and their corresponding reconstructions. Different reconstruction loss variations and latent space regularizations have been shown to improve model performances depending on the tasks to solve and to induce new desirable properties such as disentanglement. Nevertheless, measuring the success in, or enforcing properties by, the input pixel space is a challenging endeavour. In this work, we want to make use of the available data more efficiently and provide design choices to be considered in the recording or generation of future datasets to implicitly induce desirable properties during training. To this end, we propose a new sampling technique which matches semantically important parts of the image while randomizing the other parts, leading to salient feature extraction and a neglection of unimportant details. The proposed method can be combined with any existing reconstruction loss and the performance gain is superior to the triplet loss. We analyse the resulting properties on various datasets and show improvements on several computer vision tasks: illumination and unwanted features can be normalized or smoothed out and shadows are removed such that classification or other tasks work more reliably; a better invariances with respect to unwanted features is induced; the generalization capacities from synthetic to real images is improved, such that more of the semantics are preserved; uncertainty estimation is superior to Monte Carlo Dropout and an ensemble of models, particularly for datasets of higher visual complexity. Finally, classification accuracy by means of simple linear classifiers in the latent space is improved compared to the triplet loss. For each task, the improvements are highlighted on several datasets commonly used by the research community, as well as in automotive applications.

show abstract

Classifiers Based on Inverted Distances

Cited by 10 publications

References 19 publications

Entropy Based k Nearest Neighbor Pattern Classification (EbkNN): En-route to Achieving a High Accuracy in Breast Cancer Diagnosis

Entropy Based k Nearest Neighbor Pattern Classification (EbkNN): En-route to Achieving a High Accuracy in Breast Cancer Diagnosis

VOC Mixture Sensing with a MOF Film Sensor Array: Detection and Discrimination of Xylene Isomers and Their Ternary Blends

Autoencoder and Partially Impossible Reconstruction Losses

Contact Info

Product

Resources

About