-
Lessons Learned Report: Super-Resolution for Detection Tasks in Engineering Problem-Solving
Authors:
Martin Feder,
Michal Horovitz,
Assaf Chen,
Raphael Linker,
Ofer M. Shir
Abstract:
We describe the lessons learned from targeting agricultural detection problem-solving, when subject to low resolution input maps, by means of Machine Learning-based super-resolution approaches. The underlying domain is the so-called agro-detection class of problems, and the specific objective is to learn a complementary ensemble of sporadic input maps. While super-resolution algorithms are branded…
▽ More
We describe the lessons learned from targeting agricultural detection problem-solving, when subject to low resolution input maps, by means of Machine Learning-based super-resolution approaches. The underlying domain is the so-called agro-detection class of problems, and the specific objective is to learn a complementary ensemble of sporadic input maps. While super-resolution algorithms are branded with the capacity to enhance various attractive features in generic photography, we argue that they must meet certain requirements, and more importantly, that their outcome does not necessarily guarantee an improvement in engineering detection problem-solving (unlike so-called aesthetics/artistic super-resolution in ImageNet-like datasets). By presenting specific data-driven case studies, we outline a set of limitations and recommendations for deploying super-resolution algorithms for agro-detection problems. Another conclusion states that super-resolution algorithms can be used for learning missing spectral channels, and that their usage may result in some desired side-effects such as channels' synchronization.
△ Less
Submitted 1 March, 2023;
originally announced March 2023.
-
Endurance-Limited Memories: Capacity and Codes
Authors:
Yeow Meng Chee,
Michal Horovitz,
Alexander Vardy,
Van Khu Vu,
Eitan Yaakobi
Abstract:
\emph{Resistive memories}, such as \emph{phase change memories} and \emph{resistive random access memories} have attracted significant attention in recent years due to their better scalability, speed, rewritability, and yet non-volatility. However, their \emph{limited endurance} is still a major drawback that has to be improved before they can be widely adapted in large-scale systems.
In this wo…
▽ More
\emph{Resistive memories}, such as \emph{phase change memories} and \emph{resistive random access memories} have attracted significant attention in recent years due to their better scalability, speed, rewritability, and yet non-volatility. However, their \emph{limited endurance} is still a major drawback that has to be improved before they can be widely adapted in large-scale systems.
In this work, in order to reduce the wear out of the cells, we propose a new coding scheme, called \emph{endurance-limited memories} (\emph{ELM}) codes, that increases the endurance of these memories by limiting the number of cell programming operations. Namely, an \emph{$\ell$-change $t$-write ELM code} is a coding scheme that allows to write $t$ messages into some $n$ binary cells while guaranteeing that each cell is programmed at most $\ell$ times. In case $\ell=1$, these codes coincide with the well-studied \emph{write-once memory} (\emph{WOM}) codes. We study some models of these codes which depend upon whether the encoder knows on each write the number of times each cell was programmed, knows only the memory state, or even does not know anything. For the decoder, we consider these similar three cases. We fully characterize the capacity regions and the maximum sum-rates of three models where the encoder knows on each write the number of times each cell was programmed. In particular, it is shown that in these models the maximum sum-rate is $\log \sum_{i=0}^{\ell} {t \choose i}$. We also study and expose the capacity regions of the models where the decoder is informed with the number of times each cell was programmed. Finally we present the most practical model where the encoder read the memory before encoding new data and the decoder has no information about the previous states of the memory.
△ Less
Submitted 20 September, 2021;
originally announced September 2021.
-
Iterative Programming of Noisy Memory Cells
Authors:
Michal Horovitz,
Eitan Yaakobi,
Eyal En Gad,
Jehoshua Bruck
Abstract:
In this paper, we study a model, which was first presented by Bunte and Lapidoth, that mimics the programming operation of memory cells. Under this paradigm we assume that cells are programmed sequentially and individually. The programming process is modeled as transmission over a channel, while it is possible to read the cell state in order to determine its programming success, and in case of pro…
▽ More
In this paper, we study a model, which was first presented by Bunte and Lapidoth, that mimics the programming operation of memory cells. Under this paradigm we assume that cells are programmed sequentially and individually. The programming process is modeled as transmission over a channel, while it is possible to read the cell state in order to determine its programming success, and in case of programming failure, to reprogram the cell again. Reprogramming a cell can reduce the bit error rate, however this comes with the price of increasing the overall programming time and thereby affecting the writing speed of the memory. An iterative programming scheme is an algorithm which specifies the number of attempts to program each cell. Given the programming channel and constraints on the average and maximum number of attempts to program a cell, we study programming schemes which maximize the number of bits that can be reliably stored in the memory. We extend the results by Bunte and Lapidoth and study this problem when the programming channel is either the BSC, BEC, or $Z$ channel. For the BSC and the BEC our analysis is also extended for the case where the error probabilities on consecutive writes are not necessarily the same. Lastly, we also study a related model which is motivated by the synthesis process of DNA molecules.
△ Less
Submitted 10 January, 2019;
originally announced January 2019.
-
Local Rank Modulation for Flash Memories II
Authors:
Michal Horovitz,
Tuvi Etzion
Abstract:
Local rank modulation scheme was suggested recently for representing information in flash memories in order to overcome drawbacks of rank modulation. For $0 < s\leq t\leq n$ with $s$ divides $n$, an $(s,t,n)$-LRM scheme is a local rank modulation scheme where the $n$ cells are locally viewed cyclically through a sliding window of size $t$ resulting in a sequence of small permutations which require…
▽ More
Local rank modulation scheme was suggested recently for representing information in flash memories in order to overcome drawbacks of rank modulation. For $0 < s\leq t\leq n$ with $s$ divides $n$, an $(s,t,n)$-LRM scheme is a local rank modulation scheme where the $n$ cells are locally viewed cyclically through a sliding window of size $t$ resulting in a sequence of small permutations which requires less comparisons and less distinct values. The gap between two such windows equals to $s$. In this work, encoding, decoding, and asymptotic enumeration of the $(1,3,n)$-LRM scheme is studied. The techniques which are suggested have some generalizations for $(1,t,n)$-LRM, $t > 3$, but the proofs will become more complicated. The enumeration problem is presented also as a purely combinatorial problem. Finally, we prove the conjecture that the size of a constant weight $(1,2,n)$-LRM Gray code with weight two is at most $2n$.
△ Less
Submitted 20 April, 2014;
originally announced April 2014.
-
Local Rank Modulation for Flash Memories
Authors:
Michal Horovitz
Abstract:
Local rank modulation scheme was suggested recently for representing information in flash memories in order to overcome drawbacks of rank modulation. For $s\leq t\leq n$ with $s|n$, $(s,t,n)$-LRM scheme is a local rank modulation scheme where the $n$ cells are locally viewed through a sliding window of size $t$ resulting in a sequence of small permutations which requires less comparisons and less…
▽ More
Local rank modulation scheme was suggested recently for representing information in flash memories in order to overcome drawbacks of rank modulation. For $s\leq t\leq n$ with $s|n$, $(s,t,n)$-LRM scheme is a local rank modulation scheme where the $n$ cells are locally viewed through a sliding window of size $t$ resulting in a sequence of small permutations which requires less comparisons and less distinct values. The distance between two windows equals to $s$. To get the simplest hardware implementation the case of sliding window of size two was presented. Gray codes and constant weight Gray codes were presented in order to exploit the full representational power of the scheme. In this work, a tight upper-bound for cyclic constant weight Gray code in $(1,2,n)$-LRM scheme where the weight equals to $2$ is given. Encoding, decoding and enumeration of $(1,3,n)$-LRM scheme is studied.
△ Less
Submitted 19 November, 2013;
originally announced November 2013.
-
Constructions of Snake-in-the-Box Codes for Rank Modulation
Authors:
Michal Horovitz,
Tuvi Etzion
Abstract:
Snake-in-the-box code is a Gray code which is capable of detecting a single error. Gray codes are important in the context of the rank modulation scheme which was suggested recently for representing information in flash memories. For a Gray code in this scheme the codewords are permutations, two consecutive codewords are obtained by using the "push-to-the-top" operation, and the distance measure i…
▽ More
Snake-in-the-box code is a Gray code which is capable of detecting a single error. Gray codes are important in the context of the rank modulation scheme which was suggested recently for representing information in flash memories. For a Gray code in this scheme the codewords are permutations, two consecutive codewords are obtained by using the "push-to-the-top" operation, and the distance measure is defined on permutations. In this paper the Kendall's $τ$-metric is used as the distance measure. We present a general method for constructing such Gray codes. We apply the method recursively to obtain a snake of length $M_{2n+1}=((2n+1)(2n)-1)M_{2n-1}$ for permutations of $S_{2n+1}$, from a snake of length $M_{2n-1}$ for permutations of~$S_{2n-1}$. Thus, we have $\lim\limits_{n\to \infty} \frac{M_{2n+1}}{S_{2n+1}}\approx 0.4338$, improving on the previous known ratio of $\lim\limits_{n\to \infty} \frac{1}{\sqrt{πn}}$. By using the general method we also present a direct construction. This direct construction is based on necklaces and it might yield snakes of length $\frac{(2n+1)!}{2} -2n+1$ for permutations of $S_{2n+1}$. The direct construction was applied successfully for $S_7$ and $S_9$, and hence $\lim\limits_{n\to \infty} \frac{M_{2n+1}}{S_{2n+1}}\approx 0.4743$.
△ Less
Submitted 14 September, 2014; v1 submitted 19 November, 2013;
originally announced November 2013.
