More Like this

1. VehiPose: a multi-scale framework for vehicle pose estimation

   https://doi.org/10.1117/12.2595800  

   Gupta, Divyansh; Artacho, Bruno; Savakis, Andreas (July 2021, Applications of Digital Image Processing XLIV)

   Tescher, Andrew G.; Ebrahimi, Touradj (Ed.)

   Vehicle pose estimation is useful for applications such as self-driving cars, traffic monitoring, and scene analysis. Recent developments in computer vision and deep learning have achieved significant progress in human pose estimation, but little of this work has been applied to vehicle pose. We propose VehiPose, an efficient architecture for vehicle pose estimation, based on a multi-scale deep learning approach that achieves high accuracy vehicle pose estimation while maintaining manageable network complexity and modularity. The VehiPose architecture combines an encoder-decoder architecture with a waterfall atrous convolution module for multi-scale feature representation. Our approach aims to reduce the loss due to successive pooling layers and preserve the multiscale contextual and spatial information in the encoder feature representations. The waterfall module generates multiscale features, as it leverages the efficiency of progressive filtering while maintaining wider fields-of-view through the concatenation of multiple features. This multi-scale approach results in a robust vehicle pose estimation architecture that incorporates contextual information across scales and performs the localization of vehicle keypoints in an end-to-end trainable network. 

   more » « less
2. HPTMT Parallel Operators for High Performance Data Science and Data Engineering

   https://doi.org/10.3389/fdata.2021.756041  

   Abeykoon, Vibhatha; Kamburugamuve, Supun; Widanage, Chathura; Perera, Niranda; Uyar, Ahmet; Kanewala, Thejaka Amila; von Laszewski, Gregor; Fox, Geoffrey (February 2022, Frontiers in Big Data)

   Data-intensive applications are becoming commonplace in all science disciplines. They are comprised of a rich set of sub-domains such as data engineering, deep learning, and machine learning. These applications are built around efficient data abstractions and operators that suit the applications of different domains. Often lack of a clear definition of data structures and operators in the field has led to other implementations that do not work well together. The HPTMT architecture that we proposed recently, identifies a set of data structures, operators, and an execution model for creating rich data applications that links all aspects of data engineering and data science together efficiently. This paper elaborates and illustrates this architecture using an end-to-end application with deep learning and data engineering parts working together. Our analysis show that the proposed system architecture is better suited for high performance computing environments compared to the current big data processing systems. Furthermore our proposed system emphasizes the importance of efficient compact data structures such as Apache Arrow tabular data representation defined for high performance. Thus the system integration we proposed scales a sequential computation to a distributed computation retaining optimum performance along with highly usable application programming interface. 

   more » « less
3. How does Gradient Descent Learn Features---A Local Analysis for Regularized Two-Layer Neural Networks

   Zhou, Mo; Ge, Rong (December 2024, NeurIPS 2024)

   The ability of learning useful features is one of the major advantages of neural networks. Although recent works show that neural network can operate in a neural tangent kernel (NTK) regime that does not allow feature learning, many works also demonstrate the potential for neural networks to go beyond NTK regime and perform feature learning. Recently, a line of work highlighted the feature learning capabilities of the early stages of gradient-based training. In this paper we consider another mechanism for feature learning via gradient descent through a local convergence analysis. We show that once the loss is below a certain threshold, gradient descent with a carefully regularized objective will capture ground-truth directions. We further strengthen this local convergence analysis by incorporating early-stage feature learning analysis. Our results demonstrate that feature learning not only happens at the initial gradient steps, but can also occur towards the end of training. 

   more » « less
4. Robust In-Memory Computing with Hyperdimensional Stochastic Representation

   https://doi.org/10.1109/NANOARCH53687.2021.9642237  

   Poduval, P.; Issa, M.; Imani, F.; Najafi, M. H.; and Imani, M. (November 2021, Proceedings of 16th IEEE/ACM International Symposium on Nanoscale Architectures (NANOARCH 2021))

   Brain-inspired HyperDimensional Computing (HDC) is an alternative computation model working based on the observation that the human brain operates on highdimensional representations of data. Existing HDC solutions rely on expensive pre-processing algorithms for feature extraction. In this paper, we propose StocHD, a novel end-to-end hyperdimensional system that supports accurate, efficient, and robust learning over raw data. StocHD expands HDC functionality to the computing area by mathematically defining stochastic arithmetic over HDC hypervectors. StocHD enables an entire learning application (including feature extractor) to process using HDC data representation, enabling uniform, efficient, robust, and highly parallel computation. We also propose a novel fully digital and scalable Processing In-Memory (PIM) architecture that exploits the HDC memory-centric nature to support extensively parallel computation. 

   more » « less
5. StocHD: Stochastic Hyperdimensional System for Efficient and Robust Learning from Raw Data

   https://doi.org/10.1109/DAC18074.2021.9586166  

   Poduval, P; Zhou, Z.; Najafi, M. H.; Homayoun, H; and Imani, M. (December 2021, Proceedings of 58th ACM/IEEE Design Automation Conference (DAC))

   Abstract—Hyperdimensional Computing (HDC) is a neurallyinspired computation model working based on the observation that the human brain operates on high-dimensional representations of data, called hypervector. Although HDC is significantly powerful in reasoning and association of the abstract information, it is weak on features extraction from complex data such as image/video. As a result, most existing HDC solutions rely on expensive pre-processing algorithms for feature extraction. In this paper, we propose StocHD, a novel end-to-end hyperdimensional system that supports accurate, efficient, and robust learning over raw data. Unlike prior work that used HDC for learning tasks, StocHD expands HDC functionality to the computing area by mathematically defining stochastic arithmetic over HDC hypervectors. StocHD enables an entire learning application (including feature extractor) to process using HDC data representation, enabling uniform, efficient, robust, and highly parallel computation. We also propose a novel fully digital and scalable Processing In-Memory (PIM) architecture that exploits the HDC memorycentric nature to support extensively parallel computation. Our evaluation over a wide range of classification tasks shows that StocHD provides, on average, 3.3x and 6.4x (52.3x and 143.Sx) faster and higher energy efficiency as compared to state-of-the-art HDC algorithm running on PIM (NVIDIA GPU), while providing 16x higher computational robustness. 

   more » « less