Rank Code Research Articles

Deep Learning-Driven Compiler Enhancements for Efficient Matrix Multiplication

Matrix multiplication is a fundamental operation in many computational fields, requiring optimization to handle increasing data sizes efficiently. In this paper, the implementation of Deep Learning in Matrix multiplication is reviewed, which is considered important nowadays due to the growing complexity of matrix multiplication for gaming and complex programs. The current standard matrix multiplication and the time taken by it on different matrix sizes are described. The Tiled Matrix multiplication, which trims the matrix into various pieces and calculates the product for each piece, and thereafter combines the result, is also described. The times taken by both methods for different matrix sizes were compared. The main idea was to use Deep Neural Networks (DNN) to compare and rank code variants that are obtained in pieces and determine their relative performance. A tournament-based ranking system is used for assigning ranks to the code versions. The effectiveness of these techniques was evaluated on various matrix multiplication operations commonly found in deep learning workloads. Up to 8.844x speedup over the naive implementation for a matrix size of 1024 is achieved by this approach. The results demonstrate the effectiveness of combining compiler optimization techniques and deep learning models in optimizing matrix multiplication.

Query-oriented two-stage attention-based model for code search

Applying code search models to search through a large-scale codebase can significantly contribute to developers finding and reusing existing code. Researchers have applied deep learning (DL) techniques to code search models, which first compute deeper semantics representation for query and candidate code snippets, and then rank code snippets. However, these models do not well deeply analyze the semantics gap (i.e., the difference and correlation between queries written in natural language and code in programming languages), or suitably apply the correlation to the code search task. Moreover, most DL-based models use complex networks, slowing down code search tasks.To build the correlation of two languages, and apply the correlation well to code search task, we propose a query-oriented code search model named QobCS. QobCS leverage two attention-based stages, which are simple and quick, and the cooperation of the two stages bridges the semantic gap between code and query. Stage1 learns deeper semantics representation for code and query. Stage2 applies their deeper semantic correlation and query’s intention to learn better code representation.We evaluated QobCS on two datasets. On dataset1/dataset2 with 485k/542k code snippets, QobCS achieves the MRRs of 0.701/0.595, outperforming DL-based code search models DeepCS, CARLCS-CNN, UNIF, and our prior study TabCS. For efficiency, our model shows desirable performances on both datasets compared to DL-based models.

I Know What You Are Searching for: Code Snippet Recommendation from Stack Overflow Posts

Stack Overflow has been heavily used by software developers to seek programming-related information. More and more developers use Community Question and Answer forums, such as Stack Overflow, to search for code examples of how to accomplish a certain coding task. This is often considered to be more efficient than working from source documentation, tutorials, or full worked examples. However, due to the complexity of these online Question and Answer forums and the very large volume of information they contain, developers can be overwhelmed by the sheer volume of available information. This makes it hard to find and/or even be aware of the most relevant code examples to meet their needs. To alleviate this issue, in this work, we present a query-driven code recommendation tool, named Que2Code , that identifies the best code snippets for a user query from Stack Overflow posts. Our approach has two main stages: (i) semantically equivalent question retrieval and (ii) best code snippet recommendation. During the first stage, for a given query question formulated by a developer, we first generate paraphrase questions for the input query as a way of query boosting and then retrieve the relevant Stack Overflow posted questions based on these generated questions. In the second stage, we collect all of the code snippets within questions retrieved in the first stage and develop a novel scheme to rank code snippet candidates from Stack Overflow posts via pairwise comparisons. To evaluate the performance of our proposed model, we conduct a large-scale experiment to evaluate the effectiveness of the semantically equivalent question retrieval task and best code snippet recommendation task separately on Python and Java datasets in Stack Overflow. We also perform a human study to measure how real-world developers perceive the results generated by our model. Both the automatic and human evaluation results demonstrate the promising performance of our model, and we have released our code and data to assist other researchers.

An Intelligent Platform for Software Component Mining and Retrieval

The development of robotic applications necessitates the availability of useful, adaptable, and accessible programming frameworks. Robotic, IoT, and sensor-based systems open up new possibilities for the development of innovative applications, taking advantage of existing and new technologies. Despite much progress, the development of these applications remains a complex, time-consuming, and demanding activity. Development of these applications requires wide utilization of software components. In this paper, we propose a platform that efficiently searches and recommends code components for reuse. To locate and rank the source code snippets, our approach uses a machine learning approach to train the schema. Our platform uses trained schema to rank code snippets in the top k results. This platform facilitates the process of reuse by recommending suitable components for a given query. The platform provides a user-friendly interface where developers can enter queries (specifications) for code search. The evaluation shows that our platform effectively ranks the source code snippets and outperforms existing baselines. A survey is also conducted to affirm the viability of the proposed methodology.

Characterizing top ranked code examples in Google

Developers often look for code examples on the web to improve learning and accelerate development. Google indexes millions of pages with code examples: pages with better content are likely to be top ranked. In practice, many factors may influence the rank: page reputation, content quality, etc. Consequently, the most relevant information on the page, i.e., the code example, may be overshadowed by the search engine. Thus, a better understanding of how Google would rank code examples in isolation may provide the basis to detect its strengths and limitations on dealing with such content. In this paper, we assess how the Google search engine ranks code examples. We build a website with 1,000 examples and submit it to Google. After being fully indexed, we query and analyze the returned examples. We find that pages with multiple code examples are more likely to top ranked by Google. Overall, single code examples that are higher ranked are larger, however, they are not necessarily more readable and reusable. We predict top ranked examples with a good level of confidence, but generic factors have more importance than code quality ones. Based on our results, we provide insights for researchers and practitioners.

New Signature Algorithm Based on Concatenated Rank Codes

In this paper we propose a new rank code based signature scheme that used a concatenation of the LRPC and the λ-Gabidulin codes. Our construction benefits from the decoding algorithm of both of codes a considerable security levels with a moderate public key size

Infinite families of MDR cyclic codes over [formula omitted] via constacyclic codes over [formula omitted

We study α-constacyclic codes over the Frobenius non-chain ring R≔Z4[u]∕〈u2−1〉 for any unit α of R. We obtain new MDR cyclic codes over Z4 using a close connection between α-constacyclic codes over R and cyclic codes over Z4. We first explicitly determine generators of all α-constacyclic codes over R of odd length n for any unit α of R. We then explicitly obtain generators of cyclic codes over Z4 of length 2n by using a Gray map associated with the unit α. This leads to a construction of infinite families of MDR cyclic codes over Z4, where a MDR code means a maximum distance with respect to rank code in terms of the Hamming weight or the Lee weight. We obtain 202 new cyclic codes over Z4 of lengths 6, 10, 14, 18, 22, 26, 30, 34, 38, 42, 46, 50 and 54 by implementing our results in Magma software; some of them are also MDR codes with respect to the Hamming weight or the Lee weight.

New rank codes based encryption scheme using partial circulant matrices

We propose a new rank metric code based encryption based on the hard problem of rank syndrome decoding problem. We consider a generator matrix for Gabidulin codes in the form of k-partial circulant matrix. We distort the matrix G by adding it with another random k-partial circulant matrix and multiplying the product with a random circulant matrix. We also convert our encryption into an IND-CCA2 secured encryption scheme under assumption of Rank Syndrome Decoding problem. Our encryption has the smallest key size (of 4.306 KB) at 256-bit security level as compared to all the other rank code based encryption schemes with zero decryption failure and hidden structure for the decodable codes.

Effort-aware and just-in-time defect prediction with neural network.

Effort-aware just-in-time (JIT) defect prediction is to rank source code changes based on the likelihood of detects as well as the effort to inspect such changes. Accurate defect prediction algorithms help to find more defects with limited effort. To improve the accuracy of defect prediction, in this paper, we propose a deep learning based approach for effort-aware just-in-time defect prediction. The key idea of the proposed approach is that neural network and deep learning could be exploited to select useful features for defect prediction because they have been proved excellent at selecting useful features for classification and regression. First, we preprocess ten numerical metrics of code changes, and then feed them to a neural network whose output indicates how likely the code change under test contains bugs. Second, we compute the benefit cost ratio for each code change by dividing the likelihood by its size. Finally, we rank code changes according to their benefit cost ratio. Evaluation results on a well-known data set suggest that the proposed approach outperforms the state-of-the-art approaches on each of the subject projects. It improves the average recall and popt by 15.6% and 8.1%, respectively.

Application of rank metric codes in digital image watermarking

In this paper, we propose a new digital image watermarking algorithm where the resistance against attacks is studied using error correcting codes. Using the well known Lattice QIM in the spatial domain, we propose to use a different kind of error correcting codes called rank metric codes. These codes are already well used in cryptography and communications for network coding but not used yet in the context of watermarking.In this article, we show how this metric permits to correct errors with a specific structure and is adapted to specific image attacks when combined with a watermarking technique. In particular, we describe a rank metric code family called Gabidulin codes analogous to the well known Reed–Solomon codes. If one considers a rank code over a finite field extension, then any codeword has a matrix representation. One can decode the original message if the matrix rank of the detected codeword is small enough.We propose a study to validate the concept of rank metric in watermarking applications. First, we introduce a theoretically invariant method to luminance additive constant change. After combining the Lattice QIM method and rank metric codes, we add a multi-detection strategy on the damaged images with controlled luminance distortions. Then, using a block-based watermarking approach, we show how the proposed association can also be robust to an image distortion we called content erasure or copy-paste. The proposed approach completes other watermarking strategies against attacks with random errors such as JPEG compression.

The combination of sparse learning and list decoding of subspace codes for error correction in random network coding

The network coding can spread a single original error over the whole network. The simulation shows that the propagated error mostly all the time pollute just 100% of the received packets at the sink if Hamming distance is adopted. If subspace codes are adopted, usually the propagated error will not pollute 100% of the received packets in the sense of subspace distance. However, it also usually pollutes 90% of received packets which is a high error ratio. Even if the rank code and the subspace code are adopted, these existing schemes based on traditional block codes can correct corrupted errors no more than C/2 because of the limitation of the block coding where C is the max flow min cut. It is an agent to find a dense error correction method in random network coding. List decoding of subspace codes can correct frac{C}{k}hbox{-} 1 errors where k is the size of information. When frac{C}{k} is big, many errors in the sense of subspace distance can be corrected. However, the solution of list decoding is not unique. John Wright proposed a dense error correction technique based on L1 minimization, which can recover nearly 100% of the corrupted observations. In our proposal, the original packets are coded with John Wright’s coding matrix, and then, the coded message is coded again with subspace codes. In the sink, the decoding procedures about list decoding of subspace codes and John Wright’s scheme are performed. At last, the unique solution is achieved even though there are dense propagated errors in random network coding.

On the Complexity of the Rank Syndrome Decoding Problem

In this paper, we propose two new generic attacks on the rank syndrome decoding (RSD) problem. Let $C$ be a random $[n,k]$ rank code over $GF(q^{m})$ and let $y=x+e$ be a received word, such that $x \in C$ and rank $(e)=r$ . The first attack, the support attack, is combinatorial and permits to recover an error $e$ of rank weight $r$ in min $(O((n-k)^{3}m^{3}q^{r\lfloor ({km}/{n})\rfloor }, O((n-k)^{3}m^{3}q^{(r-1)\lfloor (({(k+1)m})/{n})\rfloor }))$ operations on $GF(q)$ . This new attack improves the exponent for the best generic attack for the RSD problem in the case $n > m$ , by introducing the ratio $m/n$ in the exponential coefficient of the previously best known attacks. The second attack, the annulator polynomial attack, is an algebraic attack based on the theory of $q$ -polynomials introduced by Ore. We propose a new algebraic setting for the RSD problem that permits to consider equations and unknowns in the extension field $GF(q^{m})$ rather than in $GF(q)$ as it is usually the case. We consider two approaches to solve the problem in this new setting. The linearization technique shows that if $n \ge (k+1) (r+1)-1$ the RSD problem can be solved in polynomial time. More generally, we prove that if $\lceil (({(r+1)(k+1)-} \,\, {(n+1)})/{r}) \rceil \le k$ , the RSD problem can be solved with an average complexity of $O(r^{3}k^{3}q^{r\lceil (({(r+1)(k+1)-(n+1)})/{r}) \rceil })$ operations in the base field $GF(q)$ . We also consider solving with Grobner bases for which we discuss theoretical complexity, we also consider hybrid solving with Grobner bases on practical parameters. As an example of application, we use our new attacks on all recent cryptosystems parameters, which repair the GPT cryptosystem, we break all examples of published proposed parameters, and some parameters are broken in less than 1 s in certain cases.

Learning to rank code examples for code search engines

Source code examples are used by developers to implement unfamiliar tasks by learning from existing solutions. To better support developers in finding existing solutions, code search engines are designed to locate and rank code examples relevant to user's queries. Essentially, a code search engine provides a ranking schema, which combines a set of ranking features to calculate the relevance between a query and candidate code examples. Consequently, the ranking schema places relevant code examples at the top of the result list. However, it is difficult to determine the configurations of the ranking schemas subjectively. In this paper, we propose a code example search approach that applies a machine learning technique to automatically train a ranking schema. We use the trained ranking schema to rank candidate code examples for new queries at run-time. We evaluate the ranking performance of our approach using a corpus of over 360,000 code snippets crawled from 586 open-source Android projects. The performance evaluation study shows that the learning-to-rank approach can effectively rank code examples, and outperform the existing ranking schemas by about 35.65 % and 48.42 % in terms of normalized discounted cumulative gain (NDCG) and expected reciprocal rank (ERR) measures respectively.

Information security in a random network coding network

We consider a communication network with random network coding which can be attacked by adversaries of two types. One of them can wiretap original packets outgoing from source to destination. The other can insert its own packets into information flow, which are wrong messages for the receiver. To provide secure communication, we use a scheme based on combining the GPT (Gabidulin-Paramonov-Tretjakov) public key cryptosystem and SKK (Silva-Kschischang-Koetter) codes. Encrypted packets are transmitted to the destination through wired channels. Performance of this system is investigated. The main result is that the proposed scheme is secure against wiretapping and insertion attacks under some conditions which depend on rank code parameters.

Decoding of random network codes

We consider the decoding for Silva-Kschischang-Kotter random network codes based on Gabidulin's rank-metric codes. The model of a random network coding channel can be reduced to transmitting matrices of a rank code through a channel introducing three types of additive errors. The first type is called random rank errors. To describe other types, the notions of generalized row erasures and generalized column erasures are introduced. An algorithm for simultaneous correction of rank errors and generalized erasures is presented. An example is given.

An investigation of coal-pitch interactions and their effect on thermoplastic properties at elevated pressures

The effect of pitch addition on the thermoplastic properties of two coals (British Coal rank code 301a and 502) was studied, using constant-shear-rate plastometry at elevated pressures. Increased pitch addition reduced both the plastometric ‘caking’ and ‘coking’ intensities (torques) dramatically at 2 MPa. However, whereas an increase in pitch addition increased the softening and caking temperatures of one of the coals, it had the opposite effect on the other. External gas pressure and heating rate had similar effects on the plastometric torque characteristics of a blend of coal and 7 wt% pitch to those for the coal alone. It is evident that the relations between thermoplastic properties and coal rank, pitch properties and experimental conditions such as pressure and heating rate are complex. They depend on the softening and decomposition temperatures of the coal in relation to solvation by or reaction with the pitch and the effect on the resistance to stirring in the plastometer.

An investigation of coal/pitch interactions and their effect on thermoplastic properties at elevated pressure

Abstract The effect of pitch addition on the thermoplastic properties of two coals (British Coal rank code 301a and 502) was studied, using constant-shear-rate plastometry at elevated pressures. Increased pitch addition reduced both the plastometric ‘caking’ and ‘coking’ intensities (torques) dramatically at 2 MPa. However, whereas an increase in pitch addition increased the softening and caking temperatures of one of the coals, it had the opposite effect on the other. External gas pressure and heating rate had similar effects on the plastometric torque characteristics of a blend of coal and 7 wt% pitch to those for the coal alone. It is evident that the relations between thermoplastic properties and coal rank, pitch properties and experimental conditions such as pressure and heating rate are complex. They depend on the softening and decomposition temperatures of the coal in relation to solvation by or reaction with the pitch and the effect on the resistance to stirring in the plastometer.

The determination of phenolic hydroxyl groups in coal by thermometric titration

The phenolic hydroxyl content of bituminous coals has been determined by grinding the coal with pyridine in a vibratory agate ball mill, adding acetone as a thermometric indicator and titrating with a solution of potassium hydroxide in propan-2-ol.The results indicate that phenolic hydroxyl accounts for 37 to 62 per cent. of the organic oxygen, depending on the rank of the coal. The individual values obtained are higher than those found for similar coals by potentiometric titration and by acetylation also for high carbon coals. As with these other methods, the greatest proportion of phenolic oxygen was found in coals with about 85 per cent. of carbon on the dry, mineral matter free basis (coal rank code No. 500).The conclusion is drawn that the higher values obtained by the method described are more correct because the shape of the thermometric titration curves is characteristic of those of sterically hindered phenols, many of which it is known cannot be determined by acetylation or potentiometric titration.