Deep Learning (DL) techniques are increasingly being incorporated in critical software systems today. DL software is buggy too. Recent work in SE has characterized these bugs, studied fix patterns, and proposed detection and localization strategies. In this work, we introduce a preventative measure. We propose design by contract for DL libraries, DL Contract for short, to document the properties of DL libraries and provide developers with a mechanism to identify bugs during development. While DL Contract builds on the traditional design by contract techniques, we need to address unique challenges. In particular, we need to document properties of the training process that are not visible at the functional interface of the DL libraries. To solve these problems, we have introduced mechanisms that allow developers to specify properties of the model architecture, data, and training process. We have designed and implemented DL Contract for Python-based DL libraries and used it to document the properties of Keras, a well-known DL library. We evaluate DL Contract in terms of effectiveness, runtime overhead, and usability. To evaluate the utility of DL Contract, we have developed 15 sample contracts specifically for training problems and structural bugs. We have adopted four well-vetted benchmarks from prior works on DL bug detection and repair. For the effectiveness, DL Contract correctly detects 259 bugs in 272 real-world buggy programs, from well-vetted benchmarks provided in prior work on DL bug detection and repair. We found that the DL Contract overhead is fairly minimal for the used benchmarks. Lastly, to evaluate the usability, we conducted a survey of twenty participants who have used DL Contract to find and fix bugs. The results reveal that DL Contract can be very helpful to DL application developers when debugging their code.

Automatically producing behavioral exception (BE) API documentation helps developers correctly use the libraries. The state-of-the-art approaches are either rule-based, which is too restrictive in its applicability, or deep learning (DL)-based, which requires large training dataset. To address that, we propose StatGen, a novel hybrid approach between statistical machine translation (SMT) and tree-structured translation to generate the BE documentation for any code and vice versa. We consider the documentation and source code of an API method as the two abstraction levels of the same intent. StatGen is specifically designed for this two-way inference, and takes advantage of their structures for higher accuracy.

Recently, much attentions have been paid to network coding-assisted data broadcast in vehicular networks. However, majority of the works consider all the accessed data items are the same size. In this work, we have studied the network coding-assisted heterogeneous on-demand data access in vehicular networks. Firstly, we have investigated the less efficiency of conventional coding assisted approach in accessing heterogeneous data items in real-time vehicular environment. Due to ignoring the impact of heterogeneous data items in decoding, the conventional coding does not achieve expected performance in accessing data items with diverse size. Secondly, based on our observations, we have proposed a new algorithm. The proposed network coding assisted approach exploits the different MCSs (Modulation and coding schemes) of IEEE 802.11p physical layer for leveraging the variable serving rate considering the dynamic positions' of vehicles along with the vehicle mobility. Thirdly, we have built a vehicular simulation environment and evaluated the performance of the proposed approach along with competitive no-coding and coding-assisted approaches under various circumstances. The results show that the proposed approach outperforms the state-of-the-art approaches in terms of improving the on-demand requests serving capability and reducing the data access time.

Through the embedded processors and communication technologies, vehicles are increasingly being connected with the Internet of Things. Recently, much attentions have been paid to network-coding-assisted data broadcast in vehicular networks. However, majority of the works consider all the accessed data items are the same size. In this paper, we have studied the network coding-assisted heterogeneous on-demand real-time data access in vehicular networks. First, we have investigated the less efficiency of conventional coding assisted approach in accessing heterogeneous data items in real-time vehicular environment. Due to ignoring the impact of heterogeneous data items in decoding, the conventional coding does not achieve expected performance in accessing data items with diverse size. Second, based on our observations, for efficiently serving heterogeneous data items, we have proposed a dynamic threshold-based coding-assisted real-time data broadcast approach called earliest deadline firstΘ. Third, we have derived the probabilistic analysis of the system performance of the proposed approach and the state-of-the-art approaches. Fourth, based on our further investigations, we have proposed another approach, called inverse of slack time multiply distance with THRESHOLD (Θ) (ISXDΘ). The proposed network coding-assisted ISXDΘexploits the different modulation and coding scheme of IEEE 802.11p physical layer for leveraging the variable serving rate considering the dynamic positions of vehicles along with the vehicle mobility. The comprehensive simulation results demonstrate the efficacy of the proposed approaches over the state-of-the-art approaches in terms of improving the on-demand requests serving capability and reducing the system response time.

Road Side Units (RSUs)-based data communication (V2I) is an increasingly popular technique in vehicular networks to alleviate the frequent vehicle-to-vehicle (V2V) communication problem. On-demand broadcasting is a scalable approach for supporting large clients with dynamic demands. However, the performance of the traditional on-demand broadcast can be improved significantly with the incorporation of network coding in server side broadcast. Although majority of the coding based studies assume the sizes of the server's accessed data items are the same, in reality, the sizes of the accessing data items are different. Keeping in mind of two important characteristics, from both the the vehicles' submitted requests point of view and the server's serving time point of view, we propose an efficient dynamic threshold based coding based data broadcast approach. The proposed approach considers both the deadline of request and the heterogeneity of accessed data item in scheduling, which can minimize the number of deadline missed requests and reduce the overall data access time. Simulation results verify the efficiency of our proposed approach over the traditional approaches.

Frameworks and libraries provide application programming interfaces (APIs) that serve as building blocks in modern software development. As APIs present the opportunity of increased productivity, it also calls for correct use to avoid buggy code. The usage-based specification mining technique has shown great promise in solving this problem through a data-driven approach. These techniques leverage the use of the API in large corpora to understand the recurring usages of the APIs and infer behavioral specifications (preconditions and postconditions) from such usages. A challenge for such technique is thus inference in the presence of insufficient usages, in terms of both frequency and richness. We refer to this as a "sparse usage problem." This paper presents the first technique to solve the sparse usage problem in usage-based precondition mining. Our key insight is to leverage implicit beliefs to overcome sparse usage. An implicit belief (IB) is the knowledge implicitly derived from the fact about the code. An IB about a program is known implicitly to a programmer via the language's constructs and semantics, and thus not explicitly written or specified in the code. The technical underpinnings of our new precondition mining approach include a technique to analyze the data and control flow in the program leading to API calls to infer preconditions that are implicitly present in the code corpus, a catalog of 35 code elements in total that can be used to derive implicit beliefs from a program, and empirical evaluation of all of these ideas. We have analyzed over 350 millions lines of code and 7 libraries that suffer from the sparse usage problem. Our approach realizes 6 implicit beliefs and we have observed that adding single-level context sensitivity can further improve the result of usage based precondition mining. The result shows that we achieve overall 60% in precision and 69% in recall and the accuracy is relatively improved by 32% in precision and 78% in recall compared to base usage-based mining approach for these libraries.

Road Side Units (RSUs) installed alongside the road in Vehicular Ad Hoc Networks (VANETs) act as buffer points and alleviate the frequent vehicle-to-vehicle connectivity problem. In VANETs, submitting multi-item query is a common phenomenon, for instance, a query with required traffic information of multiple routes. Unlike the single item query, a multi-item query only be satisfied successfully if all the required data items are served within the stipulated deadline. In serving multi-item query, the system also needs to address the query starvation problem which causes due to the presence of less popular data items in the same query with the high popular data items. In this paper, for serving multi-item queries efficiently, we have proposed an approach which integrates network coding with on-demand broadcasting in multi-RSU VANETs. The traditional on-demand broadcast only disseminates a single data item in a broadcast tick which restricts the maximum channel bandwidth utilization. On the contrary, our proposed approach uses network coding through which multiple data items can be broadcast in a single broadcast. Again, our proposed network coding based approach learns the cache information of vehicles intrinsically which cuts the overhead of network coding, namely avoids uploading cache information of vehicles to the RSU server explicitly. In addition, the proposed approach is equally good to integrate both with the item-level and query-level on-demand scheduling algorithms for maximizing the system performance. Finally, we have performed an extensive simulation experiment to demonstrate the superiority of our proposed approach against the traditional broadcasting system for a number of on-demand scheduling algorithms.

The usage of Road Side Units (RSUs) acting as a buffer point in VANETs alleviates the intermittent vehicle-to-vehicle (V2V) connectivity problem. However, due to the vehicle mobility and the short RSU-transmission-range, a vehicle dwells short time inside an RSU. Nevertheless, quick response from the RSU server, helps a driver to make a quick decision while driving. Network coding is known to be efficient for broadcasting multiple data items in a single packet by encoding requested data items based on the cache information of the vehicles. This can significantly improve the broadcast bandwidth usage as well as reduce the response time of the system. However, this requires the prior knowledge of cached data items of vehicles. Hence vehicles need to upload the cache information to the RSU server. This wastes upload bandwidth. In multi-RSU VANETs, we propose an approach to apply network coding in such a way so that vehicles do not need to upload their cache information to the server. Applying network coding improves the broadcast performance of an RSU in terms of minimizing the deadline miss ratio of the generated requests by vehicles, and reducing the response time of serving requests by the RSU server. Simulation results support our claims.

Network coding has become increasingly popular in time efficient data dissemination due to its inherent nature of broadcasting multiple data items in encoded form through a single broadcast. However, not much attention has been put forth on the efficiency of network coding when accessed data items' sizes can vary. In this work, we have investigated that when clients' accessed data items are of different sizes, network coding based broadcast can no longer maintain its superiority over traditional broadcast. We have found the reason of this performance degradation of coding based broadcast and propose an efficient dynamic threshold based coding approach. The proposed approach can maintain its performance superiority over both traditional broadcast approach and conventional coding based approach under a range of different size accessed data items. Simulation results support our claim.

Many safety and non-safety related applications have been envisioned in VANETs . However, efficient data dissemination considering the mobility of vehicle is must for the success of these applications. Although the Road Side Unit (RSU) is a stationary unit, both RSU and vehicle have limited transmission range that restricts to shorter connection time. This endures a higher request drop rate especially at the overloaded RSUs. A cooperative load balancing (CLB) among the RSUs to use their residual bandwidth can be an effective solution to reduce the request drop rate. In this paper, we investigate that considering the remaining delay tolerance of submitted requests and the knowledge of fixed road layout, the performance of the cooperative load balancing system can be further improved significantly. We show that this performance gain comes from serving the requests based on the urgency and the efficient load balancing among the junction-RSUs and edge-RSUs. Based on the observations, we propose an Enhanced CLB (ECLB) approach in this paper. To demonstrate the efficiency of the ECLB approach a number of well-known scheduling algorithms are integrated and an extensive simulation experiments are conducted in the vehicular communication environment that supports the superiority of ECLB over the existing approaches.