Under the auspices of the Computational Complexity Foundation (CCF)

REPORTS > KEYWORD > TENSOR PRODUCTS:
Reports tagged with Tensor Products:
TR08-105 | 26th November 2008

#### List Decoding Tensor Products and Interleaved Codes

We design the first efficient algorithms and prove new combinatorial bounds for list decoding tensor products of codes and interleaved codes.

1)We show that for every code, the ratio of its list decoding radius to its minimum distance stays unchanged under the tensor product operation (rather than squaring, as one ... more >>>

TR10-171 | 11th November 2010
Michael Viderman

#### A Note on high-rate Locally Testable Codes with sublinear query complexity

Inspired by recent construction of high-rate locally correctable codes with sublinear query complexity due to
Kopparty, Saraf and Yekhanin (2010) we address the similar question for locally testable codes (LTCs).

In this note we show a construction of high-rate LTCs with sublinear query complexity.
More formally, we show that for ... more >>>

TR11-005 | 20th January 2011

#### Testing Linear Properties: Some general themes

Revisions: 1

The last two decades have seen enormous progress in the development of sublinear-time algorithms --- i.e., algorithms that examine/reveal properties of data'' in less time than it would take to read all of the data. A large, and important, subclass of such properties turn out to be linear''. In particular, ... more >>>

TR11-087 | 3rd June 2011
Michael Viderman

#### A Combination of Testability and Decodability by Tensor Products

Revisions: 3

Ben-Sasson and Sudan (RSA 2006) showed that repeated tensor products of linear codes with a very large distance are locally testable. Due to the requirement of a very large distance the associated tensor products could be applied only over sufficiently large fields. Then Meir (SICOMP 2009) used this result (as ... more >>>

TR12-168 | 26th November 2012
Michael Viderman

#### Strong LTCs with inverse polylogarithmic rate and soundness

An error-correcting code $C \subseteq \F^n$ is called $(q,\epsilon)$-strong locally testable code (LTC) if there exists a randomized algorithm (tester) that makes at most $q$ queries to the input word. This algorithm accepts all codewords with probability 1 and rejects all non-codewords $x\notin C$ with probability at least \$\epsilon \cdot ... more >>>

ISSN 1433-8092 | Imprint