Designing Sketches for Similarity Filtering

Mic, Vladimir; Novák, David; Zezula, Pavel

doi:10.1109/icdmw.2016.0098

Cited by 11 publications

(9 citation statements)

References 21 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Lemma 2: Having two arbitrary lists of binary values I and J, e.g. values in the ith and jth bit of bit strings o ∈ X, and list ¬I of the negated values from I, the following holds 4 for the Pearson correlation coefficient corr:…”

Section: A Selecting Bits To Flipmentioning

confidence: 99%

“…Challenge 2: Having a value π = 2 l−1 , the goal is to employ the flipping of bits as defined by Equation 4 and Challenge 1 to maximize value:…”

Section: B Selecting Bits To Flip For Hwt Levelsmentioning

confidence: 99%

“…At the same time, sketches o usually have low correlated bits, as low pairwise correlations lead to good compression ratio [4]. However, these two properties cause a poor indexability [5], as the balanced bit strings have a maximum mean Hamming distance h(o 1 , o 2 ) [12] and if their bits are uncorrelated, they have the smallest variance [4]. Therefore, they have maximum distances to nearest neighbours [5].…”

Section: Bit Strings With Unbalanced Bitsmentioning

confidence: 99%

“…We binarize these datasets by the technique adopted from paper [4], which is based on generalized hyperplane partitioning (GHP) (see Figure 3). A pair of pivoting descriptors (pivots) p i1 , p i2 is selected for each bit i ∈ {0, .., λ − 1}, and value of bit o[i] expresses which of these two pivots is closer to o.…”

Section: A Testing Datamentioning

confidence: 99%

“…Many applications require efficient similarity search and volumes of current data make this task difficult to solve. Recently proposed approaches include those that substitute original data objects with bit strings [1], [2], [3], [4], [5], [6], [7]. The (dis)similarity of a pair of these bit strings is usually expressed by the Hamming distance h [8], which evaluates the number of different bits.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Modifying Hamming Spaces for Efficient Search

Mic

Novák

Zezula

2018

2018 IEEE International Conference on Data Mining Workshops (ICDMW)

Self Cite

View full text Add to dashboard Cite

We focus on the efficient search for the most similar bit strings to a given query in the Hamming space. The Hamming distance can be lower-bounded by the difference of the "number of ones" in the compared strings, i.e. of their weights. Recently, such property has been successfully used by the Hamming Weight Tree (HWT) indexing structure. We propose modifications of the bit strings that preserve pairwise Hamming distances but improve the tightness of these lower bounds, so the query evaluation with the HWT is several times faster. We also show that the unbalanced bit strings, recently reported to provide similar quality of search as the traditionally used balanced bit strings, can be more efficiently indexed with the HWT. Combined with the distance preserving modifications, the HWT query evaluation can be more than one order of magnitude faster than the HWT baseline. Keywords-Similarity search, Hamming space, HammingWeight Tree, Lower bound in the Hamming space 1 i.e. bit strings with bits set to 1 in a different ratio then a half of bit strings o ∈ X

show abstract

Section: A Selecting Bits To Flipmentioning

confidence: 99%

“…Challenge 2: Having a value π = 2 l−1 , the goal is to employ the flipping of bits as defined by Equation 4 and Challenge 1 to maximize value:…”

Section: B Selecting Bits To Flip For Hwt Levelsmentioning

confidence: 99%

Section: Bit Strings With Unbalanced Bitsmentioning

confidence: 99%

Section: A Testing Datamentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Modifying Hamming Spaces for Efficient Search

Mic

Novák

Zezula

2018

2018 IEEE International Conference on Data Mining Workshops (ICDMW)

Self Cite

View full text Add to dashboard Cite

show abstract

Metric Embedding into the Hamming Space with the n-Simplex Projection

Vadicamo

Mic

Falchi

et al. 2019

Similarity Search and Applications

Self Cite

View full text Add to dashboard Cite

Transformations of data objects into the Hamming space are often exploited to speed-up the similarity search in metric spaces. Techniques applicable in generic metric spaces require expensive learning, e.g., selection of pivoting objects. However, when searching in common Euclidean space, the best performance is usually achieved by transformations specifically designed for this space. We propose a novel transformation technique that provides a good trade-off between the applicability and the quality of the space approximation. It uses the n-Simplex projection to transform metric objects into a low-dimensional Euclidean space, and then transform this space to the Hamming space. We compare our approach theoretically and experimentally with several techniques of the metric embedding into the Hamming space. We focus on the applicability, learning cost, and the quality of search space approximation.

show abstract