To enable domain adaptation of AI on edge devices with fast convergence and low memory, we present a novel backpropagation-free optimization algorithm dubbed Target Projection Stochastic Gradient Descent (tpSGD). tpSGD uses layer-wise stochastic gradient descent (SGD) and local targets generated via random projections of the labels to train the network layer-by-layer with only forward passes. It doesn't require retaining gradients during optimization, thus greatly reducing memory allocation compared to SGD backpropagation (BP) methods. Compared to other target projection methods, tpSGD generalizes the concept to arbitrary local layer-wise loss functions. Our method performs comparably to BP gradient-descent within ~5% accuracy on relatively shallow networks of fully connected layers, convolutional layers, transformers, and recurrent layers. tpSGD also outperforms other state-of-the-art gradient-free algorithms with competitive accuracy and less memory and compute time.

Hardware Performance Counter-based (HPC) runtime checking is an effective way to identify malicious behaviors of malware and detect malicious modifications to a legitimate program’s control flow. To reduce the overhead in the monitored system which has limited storage and computing resources, we present a “sample-locally-analyze-remotely” technique. The sampled HPC data are sent to a remote server for further analysis. To minimize the I/O bandwidth required for transmission, the fine-grained HPC profiles are compressed into much smaller vectors with Compressive Sensing. The experimental results demonstrate an 80% I/O bandwidth reduction after applying Compressive Sensing, without compromising the detection and identification capabilities.

Full motion video (FMV) in unreliable, low-bit rate network channels suffers from quality issues such as jitter and block artifacts. In this paper, we introduce Vision Guided Compression (VGC), as a pre-processing technology that can be coupled with standards-based video coding, to provide FMV at low-bit rates. VGC utilizes computer vision algorithms to track salient features and keep them sharp, while non-salient features are lowpass filtered. With this approach, VGC provides an additional spatial parameter to gracefully tune the QoS, while providing FMV and preserving salient visual information.

The current state-of-the-art video coding standard, H.264, offers substantially better compression performance than earlier video coding standards. However, even with these gains, better compression is still desirable. This paper describes a macroblock-based mixed resolution video encoding system. By reducing the spatial resolution of some macroblock residuals intelligently, our preliminary experiments show up to 20% improvement in video compression efficiency compared to H.264 while maintaining good subjective video quality.

In recent decades, digital video and audio coding technologies have helped revolutionize the ways we create, deliver, and consume audiovisual content. This is exemplified by digital television (DTV), which is emerging as a captivating new program and data broadcasting service. This paper provides an overview of the video and audio coding subsystems of the Advanced Television Systems Committee (ATSC) DTV standard. We first review the motivation for data compression in digital broadcasting. The MPEG-2 video and AC-3 audio compression algorithms are described, with emphasis on basic concepts, system features, and coding performance. Next-generation video and audio codecs currently under consideration for advanced services are also presented.

In this paper, we present the work on implementation of a half-D1interlaced MPEG-4 encoder with Equator Technology DSP chip, BSP-15. The BSP-15 DSP consists mainly of a VLIW core, Co-processors, and media I/O interfaces. The encoder utilizes several BSP-15 functional blocks in parallel. In general, the VLIW performs pixel procesing that is computationally intensive. The VLx coprocessor completes variable length coding. Further parallelism is obtained by pre-loading data cache and doubling data buffers. Given the DSP processing power and real time requirements, a complexity control scheme is implemented. A frame-level quantization scheme with quality and rate control is employed. The current implementation for video at 30 fps consumes about 90% of the chip performance at a bit rate ~2Mbps.

In this paper, we propose a spatial, temporal and histogram (STH) registration algorithm for video sequences. This algorithm is developed based on a frame-level model of the misalignments often introduced by video processing algorithms, such as compression, frame rate conversion or by video capturing. With this model, the STH registration is formulated as a constrained minimization of a matching cost, and it is solved using dynamic programming. In addition, prior information about the application is incorporated in the form of contextual cost. Experimental results on both synthetic and real processed video sequences have shown that the proposed algorithm is effective and robust to a wide range of noises. In addition, we successfully applied this algorithm to digital forensic watermark detection.

This paper deals with the discrete cosine transform (DCT)-domain image resizing as compressed domain processing. Previous DCT-domain image resizing methods assume that all macroblocks are in the frame-mode. We propose an efficient method for DCT-domain image resizing with mixed field/frame-mode macroblocks. This method is developed from the investigation of the DCT-domain image resizing operation through a decomposition of matrix operations, from which we define new downsampling matrices for frame- and field-mode macroblocks. This leads to new frame- and field-mode resizing transformation matrices for DCT-domain image resizing. The proposed DCT-domain image resizing method can be used when there are mixed field/frame-mode macroblocks.