This paper develops a computationally efficient model for automatic patient-specific seizure prediction using a two-layer LSTM from multichannel intracranial electroencephalogram time-series data. We decrease the number of parameters by employing a smaller input size and fewer electrodes, thereby making the model a viable option for wearable and implantable devices. We test the proposed prediction model on 26 patients from the European iEEG dataset, which is the largest epileptic seizure dataset. We also apply an automatic preprocessing technique based on a common average reference to remove artifacts from this dataset. The simulation results show that the model with its simple structure in conjunction with the mean post-processing procedure performed the best, with an average AUC of 0.885. This study is the first that utilizes the European database for epilepsy prediction application and the first that analyzes the effect of the seizure type on the system performance and demonstrates that the seizure type has a considerable impact.

This paper presents a comprehensive analysis of hardware accelerators for neural networks in both the digital and time domains, where the latter includes spatially unrolled (SU) and recursive (REC) architectures. All accelerators are implemented and synthesized in a 65nm CMOS technology. An identical neural network model is implemented in the digital and time domain for comparative purposes in terms of throughput, power consumption, area, and energy efficiency. Post-synthesis results show that SU achieves the highest energy efficiency of 145 TOp/s/W with a throughput of 4 GOp/s. The digital core is the fastest among other cores, whereas REC is the slowest but is the most area-efficient, occupying 0.114 mm2. SU is more suited for applications with stringent power constraints and average performance, while REC is better suited for applications where the area is the most important requirement and the throughput is less significant. In contrast, the digital core is preferable for large neural networks and critical applications that require high performance.

This paper provides a comprehensive analysis of the available EEG datasets that are used for epilepsy prediction systems, including Melbourne, CHB-MIT, American Epilepsy Society, Bonn, and European Epilepsy datasets. These datasets are compared in terms of the sampling rate, number of patients, recording time, number of channels, artifacts, and types of EEG signals. We also provide details on the challenges of using one dataset over the others in predicting epilepsy. Subsequently, we compare the performance of various machine learning models that use these datasets for epileptic seizure prediction. This is the first work that provides a comprehensive analysis of various EEG datasets and should be of great importance for researchers in EEG-based systems for epileptic seizure prediction.

This article addresses the research problem of how to autonomously control Pan/Tilt/Zoom (PTZ) cameras in a manner that seeks to optimize the face recognition accuracy or the overall threat detection and proposes an overall system. The article presents two alternative schemes for camera scheduling: Grid-Based Grouping (GBG) and Elevator-Based Planning (EBP). The camera control works with realistic 3D environments and considers many factors, including the direction of the subject’s movement and its location, distances from the cameras, occlusion, overall recognition probability so far, and the expected time to leave the site, as well as the movements of cameras and their capabilities and limitations. In addition, the article utilizes clustering to group subjects, thereby enabling the system to focus on the areas that are more densely populated. Moreover, it proposes a dynamic mechanism for controlling the pre-recording time spent on running the solution. Furthermore, it develops a parallel algorithm, allowing the most time-consuming phases to be parallelized, and thus run efficiently by the centralized parallel processing subsystem. We analyze through simulation the effectiveness of the overall solution, including the clustering approach, scheduling alternatives, dynamic mechanism, and parallel implementation in terms of overall recognition probability and the running time of the solution, considering the impacts of numerous parameters.

This paper considers computer vision (CV) systems in which a central monitoring station receives and analyzes the video streams captured and delivered wirelessly by multiple cameras. It addresses how the bandwidth can be allocated to various cameras by presenting a cross-layer solution that optimizes the overall detection or recognition accuracy. In further contrast with prior work, it presents and develops a real CV system and subsequently provides a detailed experimental analysis of cross-layer optimization. Other unique features of the developed solution include employing the popular HTTP streaming approach, utilizing homogeneous cameras as well as heterogeneous ones with varying capabilities and limitations, and including a new algorithm for estimating the effective medium airtime. The results show that the proposed solution significantly improves the CV accuracy.

To fit the tight resource constraints, including network bandwidth, the video streams in Computer Vision systems are adapted dynamically by changing the video capturing and encoding parameters. We propose two novel analytical models that characterize the face recognition accuracy in terms of these parameters, specifically resolution, quantization, and actual bitrate. We find that the accuracy is a logistic function of the video quantization parameter, with the value of the Sigmoid's midpoint being a function of the resolution. Alternatively, we find that the accuracy is equal to the sum of two exponentials of the actual video bitrate, with the resolution as a multiplicative factor with one exponential. We develop an evaluation framework to validate the models using two distinct video datasets with 99 videos and the widely used Labeled Faces in the Wild (LFW) dataset with 13, 233 images. We conduct 1,668 experiments that involve varying combinations of encoding parameters. We show that both models hold true for the deep-learning and statistical-based face recognition. The developed models achieve an average coefficient of determination (R2) of 98.7% to 99.8%.

We consider the bandwidth allocation problem in automated video surveillance systems, in which a monitoring station analyzes the video streams captured and delivered wirelessly by multiple cameras. In contrast with prior studies, we provide a detailed experimental analysis of cross-layer optimization by developing a real system and conducting extensive experiments. In addition, we present an enhanced cross-layer optimization solution that allocates bandwidth to different cameras in a manner that optimizes the overall detection accuracy. The solution works with the popular HTTP streaming approach and includes a new online scheme for estimating the effective airtime of the network. The results show that the proposed solution significantly improves the detection accuracy.

This article develops an aggregate power consumption model for live video streaming systems, including many-to-many systems. In many-to-one streaming systems, multiple video sources (i.e., cameras and/or sensors) stream videos to a monitoring station. We model the power consumed by the video sources in the capturing, encoding, and transmission phases and then provide an overall model in terms of the main capturing and encoding parameters, including resolution, frame rate, number of reference frames, motion estimation range, and quantization. We also analyze the power consumed by the monitoring station due to receiving, decoding, and upscaling the received video streams. In addition to modeling the power consumption, we model the achieved bitrate of video encoding. We validate the developed models through extensive experiments using two types of systems and different video contents. Furthermore, we analyze many-to-one systems in terms of bitrate, video quality, and the power consumed by the sources, as well as that by the monitoring station, considering the impacts of multiple parameters simultaneously.

The design of interactive Near Video-on-Demand (NVOD) systems is highly complicated when scalable stream merging is used. We propose an intelligent client-side cache management policy for these systems, allowing and exploiting cache discontinuity. This policy maximizes the percentage of interactive requests serviced from the client's own cache without requiring any resources from the server. The policy caches data from all streams that are being listened to by the client. As the cache becomes full, it purges data according to a purging algorithm. We present three purging algorithms: purge oldest data, purge the furthest data from the customer's playback point, and purge adaptively. Moreover, we experiment with another important decision, which is whether pausing users should continue to listen to streams when the cache becomes full. We evaluate the effectiveness of the proposed cache management policy and purging algorithms under realistic and complex workload through extensive simulations. We analyze many metrics, including waiting and blocking metrics, aggregate delay, cache hit rate, and cache fragmentation, considering a variety of system parameters.

The efficient control of Pan/Tilt/Zoom (PTZ) cameras has been a major research problem. This paper presents a solution that seeks to optimize the overall subject recognition probability by controlling various deployed cameras, based on the characteristics of the subjects in the surveillance area. In particular, we propose and analyze a clustering-based approach, which can be used in conjunction with recently proposed camera scheduling schemes, to achieve significant improvements in both the subject recognition probability and the algorithm computation time. We extensively analyze the effectiveness of the clustering approach, considering the impacts of subject arrival rate.

This paper develops an accuracy-based cross-layer optimization solution for wireless automated video surveillance systems, in which multiple sources stream videos to a central proxy station. The proposed solution manages the application rates and transmission opportunities of various video sources based on the dynamic network conditions in such a way that maximizes the overall detection accuracy of the computer vision algorithm(s). We demonstrate the effectiveness of the proposed solution through extensive experiments.

The required real-time and high-rate transfer of multimedia data limits the numbers of requests that can be concurrently serviced by video-on-demand (VOD) systems. Resource-sharing techniques can be used to address this scalability challenge, but they greatly complicate the efficient support for interactive operations. We develop an overall solution for interactive near VOD systems that employ resource sharing. The proposed solution supports user interactions with short response times and low rejection probabilities. The solution includes a novel stream provisioning policy, which dynamically determines the best number of I-Streams (unicast streams for supporting interactive requests) and the maximum I-Stream length that can be allocated by the server. Furthermore, we use a sophisticated client-side cache management policy to maximize the percentage of interactive requests serviced from the client's own cache. We study the system using realistic workload through extensive simulation.

The design of interactive Video-on-Demand (VOD) systems is highly complicated when scalable stream merging is used. Support of requests under the Near Video-on-Demand (NVOD), where not all requests are serviced immediately, introduces additional complications. We develop an overall solution for designing interactive NVOD systems when stream merging is employed. The proposed solution supports user interactions with short response times and low rejection probabilities. In addition, we propose an efficient dynamic stream allocation policy and use a sophisticated client-side cache management policy. We analyze system performance, through extensive simulation, using realistic workload under different levels of user interactivity.

Power consumption of video streaming systems has become a major concern, especially in battery-powered devices, such as video sensors. Power is usually dissipated in each one of the major phases of the streaming process: capturing, encoding, and transmission. This paper develops models for power consumption in each of these phases and validates them with extensive experiments, focusing primarily on H.264 video encoding. For comparative purposes, we also study MJPEG and MPEG-4 video codecs. In addition, we analyze the impacts of the main H.264 video compression parameters on power consumption and bitrate. These parameters include quantization parameter, number of reference frames, motion estimation (ME) range, and ME algorithm.

This paper develops a cross-layer optimization framework for video streaming from multiple sources to a central proxy station over a wireless network. The proposed framework manages the application rates and transmission opportunities of various video sources based on the dynamic network conditions in such a way that minimizes the overall distortion. The framework utilizes a novel online approach for estimating the effective airtime of the network. We demonstrate the effectiveness of the proposed framework and effective airtime estimation approach through extensive experiments.

This paper analyzes and compares the rate-accuracy and rate energy characteristics of various video rate adaptation techniques in computer vision applications. The analyzed rate adaptation techniques include spatial, spatial with up scaling, temporal, and Signal-to-Noise Ratio (SNR). We experiment with standard video sequences as well as 300 security, surveillance, news, and speech videos. These videos total 19.15 hours of recording time. We consider both MPEG-4 and H.264 compression standards.

This paper deals with the camera control problem in automated video surveillance. We develop a solution that seeks to optimize the overall subject recognition probability by controlling the pan, tilt, and zoom of various deployed Pan/Tilt/Zoom (PTZ) cameras. Since the number of subjects is usually much larger than the number of video cameras, the problem to be addressed is how to assign subjects to these cameras. This control of cameras is based on the direction of the subject's movement and its location, distances from the cameras, occlusion, overall recognition probability so far, and the expected time to leave the site, as well as the movements of cameras and their capabilities and limitations. The developed solution works with realistic 3D environments and not just 2D scenes. We analyze the effectiveness of the proposed solution through extensive simulation.

VP8 has recently been offered by Google as an open video compression format in attempt to compete with the widely used H.264 video compression standard. This paper describes the major differences between VP8 and H.264 and provides detailed comparative evaluations through extensive experiments. We use 29 raw video sequences, offering a wide spectrum of resolutions and content characteristics, with the resolution ranging from 176×144 (QCIF) to 3840×2160 (2160p). To ensure a fair study, we use 3 coding presets in H.264, each with three types of tuning, and 7 presets in VP8. The presets cover a variety of achieved quality or complexity levels. The performance metrics include accuracy of bit rate handling, encoding speed, decoding speed, and perceptual video quality.

This paper considers scalable delivery of streaming video content with advertisements. It proposes a highly accurate waiting-time prediction algorithm that estimates the expected number of viewed ads by utilizing detailed information about the system state and the applied scheduling policy. It also proposes a pricing scheme based on the expected waiting times, which include the ads' viewing times. The revenues generated by the ads are used to subsidize the price and are allocated to clients proportionally to their expected waiting times. The envisioned system presents the client with an updated menu of the supported videos (such as full-length featured movies) along with the corresponding ads' viewing times and prices. We analyze the effectiveness of the proposed prediction algorithm and pricing scheme in terms of many metrics, considering numerous design and workload parameters, and capturing purchasing capacity and willingness models.

The number of video streams that can be serviced concurrently is highly constrained by the required real-time and high-rate transfers of multimedia data. Resource sharing techniques, such as Batching, Patching, and Earliest Reachable Merge Target (ERMT), can be used to address this problem by utilizing the multicast facility, which allows multiple requests to share the same set of server and network resources. They assume, however, that all clients have the same available download bandwidth and buffer space. We study how to efficiently support clients with varying available download bandwidth and buffer space, while delivering data in a client-pull fashion using enhanced resource sharing. In particular, we propose three hybrid solutions to address the variability in the download bandwidth among clients: Simple Hybrid Solution (SHS), Adaptive Hybrid Solution (AHS), and Enhanced Hybrid Solution (EHS). SHS simply combines Batching with either Patching or ERMT, leading to two alternatives: SHS-P and SHS-E , respectively. Batching is used for clients with bandwidth lower than double the video playback rate, and Patching/ERMT is used for the rest. In contrast, AHS and EHS classify clients into multiple bandwidth classes and service them accordingly. AHS employs a new stream type, called adaptive stream , and EHS employs an enhanced adaptive stream type to serve clients with bandwidth capacities ranging between the video playback rate and double that rate. AHS and EHS employ adaptive streams or enhanced adaptive streams in conjunction with Batching and Patching or ERMT, leading to four possible schemes: AHS-P, AHS-E, EHS-P, and EHS-E. Moreover, we consider the variability of the available buffer space among clients. Furthermore, we study how the waiting playback requests for different videos can be scheduled for service in the heterogeneous environment, capturing the variations in both the client bandwidth and buffer space. We evaluate the effectiveness of the proposed solutions and analyze various scheduling policies through extensive simulation.

Providing video streaming users with expected waiting times enhances their perceived quality-of-service (QoS) and encourages them to wait. In the absence of any waiting-time feedback, users are more likely to defect because of the uncertainty as to when their services will start. We analyze waiting-time predictability in scalable video streaming. We propose two prediction schemes and study their effectiveness when applied with various stream merging techniques and scheduling policies. The results demonstrate that the waiting time can be predicted accurately, especially when enhanced cost-based scheduling is applied. The combination of waiting-time prediction and cost-based scheduling leads to outstanding performance benefits.

The required real-time and high-rate transfers for multimedia data severely limit the number of video streams that can be delivered concurrently. Resource-sharing techniques address this problem and can be classified into two main classes: stream merging and periodic broadcasting. We evaluate through extensive simulation major resource-sharing techniques from the two classes, considering different service models and video workloads. We utilize this extensive analysis in developing a workload-aware hybrid solution (WAHS) that combines the advantages of the best performers among resource-sharing techniques. Moreover, we propose a statistical cache management (SCM) approach and derive analytical models for optimal cache allocation to reduce further the demands on the disk I/O when various resource sharing techniques are used.

We propose a scheduling policy, called Predictive Cost-Based Scheduling (PCS) for scalable video streaming. PCS schedules the waiting requests based on their required delivery costs, predicts future system state, and uses the prediction results to possibly alter the scheduling decisions. We also present two alternative implementations of PCS and analyze its waiting-time predictability. The simulation results show that PCS can achieve significant performance improvements and can provide users with highly accurate expected waiting times.

Recent workload characterization studies show that a large number of video streaming sessions do not access videos in a sequential manner from the beginning to the end, as assumed in most prior work. In this paper, we study the impact of realistic access patterns on resource sharing and propose and evaluate five enhancements to reduce server load and improve customer-perceived quality-of-service (QoS).

This paper presents a delivery framework for streaming media with advertisements and an associated pricing model. The delivery model combines the benefits of periodic broadcasting and stream merging. The advertisements' revenues are used to subsidize the price of the media content. The pricing is determined based on the total ads' viewing time. Moreover, this paper presents three modified scheduling policies that are well suited to the proposed delivery framework and analyzes their effectiveness through simulation.

Providing video streaming users with expected waiting times enhances their perceived quality-of-service (QoS) and encourages them to wait. In the absence of any waiting-time feedback, users are more likely to defect because of the uncertainty as to when they will start to receive services. In this paper, we analyze waiting-time predictability in scalable video streaming. We present three prediction schemes and study their effectiveness when applied with various stream merging techniques and scheduling policies. The results demonstrate that the waiting time can be predicted accurately, especially when enhanced cost-based scheduling is applied. The combination of waiting-time prediction and cost-based scheduling leads to outstanding performance benefits.

The required real-time and high-rate transfers for multimedia data severely limit the number of requests that can be serviced by Video-on-Demand (VOD) servers. Resource sharing techniques can be used to address this problem. We evaluate through extensive simulation major resource sharing techniques, considering both the True Video-on- Demand (TVOD) and Near Video-on-Demand (NVOD) service models. Moreover, we propose a statistical approach for cache management and derive analytical models for optimal cache allocation to reduce the demands on the disk I/O when various resource sharing techniques are used.

The required real-time and high-rate transfers for multimedia data severely limit the number of requests that can be serviced concurrently by Video-on-Demand (VOD) servers. Resource sharing techniques can be used to address this problem. We study how VOD servers can support heterogeneous receivers while delivering data in a client-pull fashion using enhanced resource sharing. We propose three hybrid solutions. The first solution simply combines existing resource sharing techniques and deals with clients as two bandwidth classes. The other two solutions, however, classify clients into multiple bandwidth classes and service them accordingly by capturing the proposed ideas of Adaptive Stream Merging or Enhanced Adaptive Stream Merging, respectively. We also discuss how scheduling policies can be adapted to the heterogeneous environment so as to exploit the variations in client bandwidth. We evaluate the effectiveness of the proposed solutions and analyze various scheduling policies through extensive simulation.

Multimedia-on-demand (MOD) applications have grown dramatically in popularity, especially in the domains of education, business, and entertainment. Current MOD servers waste precious resources in performing store-and-forward copying. This excessive overhead increases cost and severely limits the scalability of these servers. In this paper, we propose using the network-attached disk (NAD) architecture to design highly scalable and cost-effective MOD servers. In order to ensure enhanced performance, we propose a scheme, called distributed interval caching (DIG), which utilizes the on-disk buffers for caching intervals between successive streams. We also propose another scheme, called multiobjective scheduling (MOS), which increases the degrees of resource sharing by scheduling the waiting requests for service intelligently. We then integrate the two schemes and study the overall performance benefits through extensive simulation. The results demonstrate that the integrated policy works very well in increasing the number of customers that can be serviced concurrently while decreasing their waiting times for service. The performance benefits vary with several architectural, system workload, and scheduling parameters. We conclude this study by developing an analytical model for ideal DIG in order to estimate the performance limits which may be achieved through various optimizations.

In this paper we propose using the network-attached disk (NAD) architecture to design highly scalable and cost-effective multimedia-on-demand (MOD) servers. In order to ensure enhanced performance, we propose two schemes, called distributed interval caching (DIC) and multi-objective scheduling (MOS). The DIC scheme utilizes the on-disk buffers for caching intervals between successive streams, while the MOS scheme improves resource sharing by scheduling requests for service intelligently based on four predefined criteria. We then integrate the two schemes and study the overall performance benefits through extensive simulation. We also study the effectiveness of the proposed DIC scheme by developing an analytical model that estimates the performance limit of DIC The results demonstrate that the integrated policy works very well in increasing the number of customers that can be serviced concurrently while decreasing their waiting times, and that the performance improvements scale with the number of disks in the server.

Video-on-demand (VOD) is increasingly becoming one of the most important and successful services due to the recent advances in storage subsystems, compression technology and, networking. Therefore, the investigation of various alternatives to improve the performance of VOD servers has become a major research focus. The reduction of disk access time through intelligent data placement strategies is one such avenue and is the theme of this paper: Movie rental patterns indicate that accesses to movies are highly localized with only a small number of movies receiving most of the accesses. In this paper we exploit the access patterns and propose an adaptive rearrangement of the blocks on each disk within the server. With this approach, the blocks of the movies with comparable access frequencies are kept closer to each other We analyze two rearrangement schemes, called centered and sequential. In the centered layout, blocks are placed according to their access patterns starting with the most popular movie at the center. The sequential layout places movies in the order of their popularity starting at the edge of the disk. We compare and evaluate, through an intensive simulation study, the effectiveness of these layouts with respect to arbitrary layouts. The simulation results indicate that significant disk improvements could be attained by adopting the proposed schemes, and that the centered layout is the best performer.