ABSTRACT Emotion detection from face and speech is inherent for human–computer interaction, mental health assessment, social robotics, and emotional intelligence. Traditional machine learning methods typically depend on handcrafted features and are primarily centred on unimodal systems. However, the unique characteristics of facial expressions and the variability in speech features present challenges in capturing complex emotional states. Accordingly, deep learning models have been substantial in automatically extracting intrinsic emotional features with greater accuracy across multiple modalities. The proposed article presents a comprehensive review of recent progress in emotion detection, spanning from unimodal to multimodal systems, with a focus on facial and speech modalities. It examines state‐of‐the‐art machine learning, deep learning, and the latest transformer‐based approaches for emotion detection. The review aims to provide an in‐depth analysis of both unimodal and multimodal emotion detection techniques, highlighting their limitations, popular datasets, challenges, and the best‐performing models. Such analysis aids researchers in judicious selection of the most appropriate dataset and audio‐visual emotion detection models. Key findings suggest that integrating multimodal data significantly improves emotion recognition, particularly when utilising deep learning methods trained on synchronised audio and video datasets. By assessing recent advancements and current challenges, this article serves as a fundamental resource for researchers and practitioners in the field of emotional AI, thereby aiding in the creation of more intuitive and empathetic technologies.

In the last few decades, technology advancements have paved the way for the creation of intelligent and autonomous systems that utilize complex calculations which are both time-consuming and central processing unit intensive. As a consequence, parallel processing systems are gaining popularity to enhance overall computer performance. Programmers should be able to efficiently utilize available hardware resources with parallelization in an ideal world. Through the automatic parallelization of sequential code, multithreading can be executed without extra supervision. However, a wide range of software dependencies prevents this from being feasible. An architectural framework for speculative parallelization along with an efficient memory analysis and computational algorithms for the code generation are proposed that can provide optimal performance. Furthermore, a suitable support of hardware design as a runtime library to the proposed architectural framework is presented which can be used to recover misspeculated results during execution to minimize speculative parallelism overhead. The implementation makes use of the Low-Level Virtual Machine compiler infrastructure and is tested on numerous benchmarks, thus making it highly scalable in terms of programming languages and architectures. According to our experimental results, there is significant potential for speedup increase. In comparison to the overall function speedup, that is, geomean speedup of 5.2× approximately when using the proposed architecture without hardware support, the proposed architectural framework and algorithm with hardware support give an average geomean speedup of 7.0× approximately on the given benchmark which is written in C/C++.

Summary Information storage and retrieval from the cloud is growing continuously unabated. The cost‐efficient solutions offered by the cloud providers to the end‐users have motivated them to outsource their confidential data to the cloud. Outsourcing confidential data leads to enhanced privacy risks due to disclosure of sensitive information to adversaries. To handle this disclosure of information, encryption is preferred, but it hinders the efficient searching on the documents. The existing searchable encryption schemes either focused on optimization of search time or improvement of search efficiency. To solve this trade‐off between search time and search efficiency, we propose an efficient disjunctive search scheme using non‐positional inverted index. To the best of our knowledge, there is no searchable encryption scheme based on the non‐positional inverted index in the literature. Thus, we first propose a basic scheme based on a non‐positional inverted index to search the desired keywords and highlight its inefficiency in terms of high search time required. To perform efficient searching, an extended search scheme is proposed using keyword binning, which reduces comparisons required and improves the search time. The extended search scheme has recall of 100% and precision 99.75%. The experimental analysis of the proposed scheme on real datasets proves that the proposed scheme is privacy‐preserving and efficient.

Image denoising is a thoroughly studied research problem in the areas of image processing and computer vision. In this work, a deep convolution neural network with added benefits of residual learning for image denoising is proposed. The network is composed of convolution layers and ResNet blocks along with rectified linear unit activation functions. The network is capable of learning end‐to‐end mappings from noise distorted images to restored cleaner versions. The deeper networks tend to be challenging to train and often are posed with the problem of vanishing gradients. The residual learning and orthogonal kernel initialisation keep the gradients in check. The skip connections in the ResNet blocks pass on the learned abstractions further down the network in the forward pass, thus achieving better results. With a single model, one can tackle different levels of Gaussian noise efficiently. The experiments conducted on the benchmark datasets prove that the proposed model obtains a significant improvement in structural similarity index than the previously existing state‐of‐the‐art techniques.

Traffic congestion on urban roadways is a serious problem requiring novel ways to detect and mitigate it. Determining the routes that lead to the traffic congestion segment is also vital in devising mitigation strategies. Further, crowdsourcing this information allows for use of these strategies quickly and in places where infrastructure is not available. In this work, we present an unconventional method, using the barometer sensor of mobile phones to (a) detect road traffic congestion and (b) estimate the paths that lead to the congested road segment. We make the observation that roads are not completely flat and very often, altitude varies along the road. The barometer sensor chips are sensitive enough to measure these variations and consume very little energy of the mobile phone, compared to other sensors such as the GPS or accelerometer. We devise a feature set to map the rate of change of this altitude as the user moves into activities characterized as “still” and “motion,” which are further used by the traffic congestion detection algorithm (RoadSphygmo) to classify the group of users as being in “moving,” “congestion,” or “stuck” states. To estimate the paths that lead to the congested road segment, we compare the user’s barometer sensor readings with a pre-stored road signature of barometer values using Dynamic Time Warping (DTW). We show that by using correlation of barometer sensor values, we can determine if users are in the same vehicle. We crowdsource this information from multiple mobile phones and use majority voting technique to improve the accuracy of traffic congestion detection and path estimation. We find a significant increase in the accuracies using crowdsourced information as compared to individual mobile phones. Further, we show that we can use barometer sensor for other applications such as bus occupancy, boarding/deboarding of a vehicle, and so on. The validation of the state determined by RoadSphygmo is done by comparing it with average GPS speed calculated during the same time period. The path estimation is validated over different intersections and considering various cases of commuter travel. The results obtained are promising and show that the traffic state determination and the estimation of the path taken by the commuter can achieve high accuracy.

Summary Organizations prefer using cloud for storing their data due to availability of cost‐effective storage. The outsourced data include sensitive information, so data is encrypted as maintaining confidentiality and privacy of the documents is of paramount importance. Retrieving the desired information from the cloud requires efficient searching, which involves submission of search query to the cloud server by the end‐user. As the search terms may include sensitive information of an organization, it is desired that the search query should not reveal any confidential information. The existing works are not suitable for big‐data scenario due to high search time required for large document collections, thereby leading to increased cloud usage cost. Thus, an efficient approach to perform search on encrypted data using clustering is proposed in this paper. As the proposed technique clusters the documents based on the relationship between the keywords, the search method involves searching documents within the relevant cluster in contrast to searching the entire dataset. An efficient ranking method is incorporated to rank the documents according to the relevance to search query using Term Frequency‐Inverse Document Frequency (tf‐idf) value of the keywords in the documents, which leads to reduced communication overheads due to reduction in unnecessary documents being downloaded. Moreover, an efficient query randomization approach is proposed so that two or more queries involving the same search terms appear distinct. Experimental results using real datasets demonstrate that our proposed multi‐keyword ranked search scheme on encrypted cloud data significantly reduce the number of comparisons and search time in comparison to the existing techniques while maintaining recall of 100% and precision of 82%.

Summary Outsourcing confidential data to cloud storage leads to privacy challenges that can be reduced using encryption. However, with encryption in place, the utilization of the data is reduced, which leads to reduced quality of experience of the users. To overcome this, searchable encryption (SE) schemes are utilized, which allow the end users to retrieve the relevant documents from the cloud, for which various researchers have worked utilizing different techniques. Despite the popularity of the searchable encryption schemes, most of the surveys either do not provide or present an incomplete taxonomy of SE schemes. Hence, in this paper, we attempt to present a complete taxonomy/classification of the searchable encryption schemes in terms of the type of search, type of index, results retrieved, implementation type, multiplicity of users, and the technique used. From the literature, it is observed that inner product similarity is widely adopted by researchers to compute the similarity of the query and the document index as it provides both conjunctive and disjunctive searching (ie, have better search capability) but requires high search time (ie, have lower search efficiency). On the other hand, schemes based on binary comparisons exist, which require less search time (ie, have better search efficiency) but support only conjunctive searching (ie, have limited search capability). Thus, a major conclusion drawn from our work is that there is an imbalance between search capability and search efficiency, ie, in the existing schemes, search capability can be improved at the cost of search time only. Therefore, we suggest that one direction where researchers should work on is to provide a balance between search capability and search efficiency.

Determining the location of a mobile user is central to several crowd-sensing applications. Using a Global Positioning System is not only power-hungry, but also unavailable in many locations. While there has been work on cellular-based localization, we consider an unexplored opportunity to improve location accuracy by combining cellular information across multiple mobile devices located near each other. For instance, this opportunity may arise in the context of public transport units having multiple travelers. Based on theoretical analysis and an extensive experimental study on several public transportation routes in two cities, we show that combining cellular information across nearby phones considerably improves location accuracy. Combining information across phones is especially useful when a phone has to use another phone’s fingerprint database, in a fingerprinting-based localization scheme. Both the median and 90 percentile errors reduce significantly. The location accuracy also improves irrespective of whether we combine information across phones connected to the same or different cellular operators. Sharing information across phones can raise privacy concerns. To address this, we have developed an id-free broadcast mechanism, using audio as a medium, to share information among mobile phones. We show that such communication can work effectively on smartphones, even in real-life, noisy-road conditions.