-
PikeLPN: Mitigating Overlooked Inefficiencies of Low-Precision Neural Networks
Authors:
Marina Neseem,
Conor McCullough,
Randy Hsin,
Chas Leichner,
Shan Li,
In Suk Chong,
Andrew G. Howard,
Lukasz Lew,
Sherief Reda,
Ville-Mikko Rautio,
Daniele Moro
Abstract:
Low-precision quantization is recognized for its efficacy in neural network optimization. Our analysis reveals that non-quantized elementwise operations which are prevalent in layers such as parameterized activation functions, batch normalization, and quantization scaling dominate the inference cost of low-precision models. These non-quantized elementwise operations are commonly overlooked in SOTA…
▽ More
Low-precision quantization is recognized for its efficacy in neural network optimization. Our analysis reveals that non-quantized elementwise operations which are prevalent in layers such as parameterized activation functions, batch normalization, and quantization scaling dominate the inference cost of low-precision models. These non-quantized elementwise operations are commonly overlooked in SOTA efficiency metrics such as Arithmetic Computation Effort (ACE). In this paper, we propose ACEv2 - an extended version of ACE which offers a better alignment with the inference cost of quantized models and their energy consumption on ML hardware. Moreover, we introduce PikeLPN, a model that addresses these efficiency issues by applying quantization to both elementwise operations and multiply-accumulate operations. In particular, we present a novel quantization technique for batch normalization layers named QuantNorm which allows for quantizing the batch normalization parameters without compromising the model performance. Additionally, we propose applying Double Quantization where the quantization scaling parameters are quantized. Furthermore, we recognize and resolve the issue of distribution mismatch in Separable Convolution layers by introducing Distribution-Heterogeneous Quantization which enables quantizing them to low-precision. PikeLPN achieves Pareto-optimality in efficiency-accuracy trade-off with up to 3X efficiency improvement compared to SOTA low-precision models.
△ Less
Submitted 29 March, 2024;
originally announced April 2024.
-
SAMVG: A Multi-stage Image Vectorization Model with the Segment-Anything Model
Authors:
Haokun Zhu,
Juang Ian Chong,
Teng Hu,
Ran Yi,
Yu-Kun Lai,
Paul L. Rosin
Abstract:
Vector graphics are widely used in graphical designs and have received more and more attention. However, unlike raster images which can be easily obtained, acquiring high-quality vector graphics, typically through automatically converting from raster images remains a significant challenge, especially for more complex images such as photos or artworks. In this paper, we propose SAMVG, a multi-stage…
▽ More
Vector graphics are widely used in graphical designs and have received more and more attention. However, unlike raster images which can be easily obtained, acquiring high-quality vector graphics, typically through automatically converting from raster images remains a significant challenge, especially for more complex images such as photos or artworks. In this paper, we propose SAMVG, a multi-stage model to vectorize raster images into SVG (Scalable Vector Graphics). Firstly, SAMVG uses general image segmentation provided by the Segment-Anything Model and uses a novel filtering method to identify the best dense segmentation map for the entire image. Secondly, SAMVG then identifies missing components and adds more detailed components to the SVG. Through a series of extensive experiments, we demonstrate that SAMVG can produce high quality SVGs in any domain while requiring less computation time and complexity compared to previous state-of-the-art methods.
△ Less
Submitted 25 December, 2023; v1 submitted 9 November, 2023;
originally announced November 2023.
-
Microservices based Framework to Support Interoperable IoT Applications for Enhanced Data Analytics
Authors:
Sajjad Ali,
Muhammad Aslam Jarwar,
Ilyoung Chong
Abstract:
Internet of things is growing with a large number of diverse objects which generate billions of data streams by sensing, actuating and communicating. Management of heterogeneous IoT objects with existing approaches and processing of myriads of data from these objects using monolithic services have become major challenges in developing effective IoT applications. The heterogeneity can be resolved b…
▽ More
Internet of things is growing with a large number of diverse objects which generate billions of data streams by sensing, actuating and communicating. Management of heterogeneous IoT objects with existing approaches and processing of myriads of data from these objects using monolithic services have become major challenges in developing effective IoT applications. The heterogeneity can be resolved by providing interoperability with semantic virtualization of objects. Moreover, monolithic services can be substituted with modular microservices. This article presents an architecture that enables the development of IoT applications using semantically interoperable microservices and virtual objects. The proposed framework supports analytic features with knowledge-driven and data-driven techniques to provision intelligent services on top of interoperable microservices in Web Objects enabled IoT environment. The knowledge-driven aspects are supported with reasoning on semantic ontology models and the data-driven aspects are realized with machine learning pipeline. The development of service functionalities is supported with microservices to enhance modularity and reusability. To evaluate the proposed framework a proof of concept implementation with a use case is discussed.
△ Less
Submitted 19 October, 2019;
originally announced October 2019.
-
Microservices based Linked Data Quality Model for Buildings Energy Management Services
Authors:
Muhammad Aslam Jarwar,
Sajjad Ali,
Ilyoung Chong
Abstract:
During the production, distribution, and consumption of energy, a large quantity of data is generated. For efficiently using of energy resources other supplementary data such as building information, weather, and environmental data etc. are also collected and used. All these energy data and relevant data is published as linked data in order to enhance the reusability of data and maximization of en…
▽ More
During the production, distribution, and consumption of energy, a large quantity of data is generated. For efficiently using of energy resources other supplementary data such as building information, weather, and environmental data etc. are also collected and used. All these energy data and relevant data is published as linked data in order to enhance the reusability of data and maximization of energy management services capability. However, the quality of this linked data is questionable because of wear and tears of sensors, unreliable communication channels, and highly diversification of data sources. The provision of high-quality energy management services requires high quality linked data, which reduces billing cost and improve the quality of the living environment. Assessment and improvement methodologies for the quality of data along with linked data needs to process very diverse data from highly diverse data sources. Microservices based data-driven architecture has great significance to processes highly diverse linked data with modularity, scalability, and reliability. This paper proposed microservices based architecture along with domain data and metadata ontologies to enhance and assess energy-related linked data quality.
△ Less
Submitted 11 October, 2019;
originally announced October 2019.
-
On the determination of the Fermi surface in high-Tc superconductors by angle-resolved photoemission spectroscopy
Authors:
J. Mesot,
M. Randeria,
M. R. Norman,
A. Kaminski,
H. M. Fretwell,
J. C. Campuzano,
H. Ding,
T. Takeuchi,
T. Sato,
T. Yokoya,
T. Takahashi,
I. Chong,
T. Terashima,
M. Takano,
T. Mochiku,
K. Kadowaki
Abstract:
We study the normal state electronic excitations probed by angle resolved photoemission spectroscopy (ARPES) in Bi2201 and Bi2212. Our main goal is to establish explicit criteria for determining the Fermi surface from ARPES data on strongly interacting systems where sharply defined quasiparticles do not exist and the dispersion is very weak in parts of the Brillouin zone. Additional complication…
▽ More
We study the normal state electronic excitations probed by angle resolved photoemission spectroscopy (ARPES) in Bi2201 and Bi2212. Our main goal is to establish explicit criteria for determining the Fermi surface from ARPES data on strongly interacting systems where sharply defined quasiparticles do not exist and the dispersion is very weak in parts of the Brillouin zone. Additional complications arise from strong matrix element variations within the zone. We present detailed results as a function of incident photon energy, and show simple experimental tests to distinguish between an intensity drop due to matrix element effects and spectral weight loss due to a Fermi crossing. We reiterate the use of polarization selection rules in disentangling the effect of umklapps due to the BiO superlattice in Bi2212. We conclude that, despite all the complications, the Fermi surface can be determined unambiguously: it is a single large hole barrel centered about (pi,pi) in both materials.
△ Less
Submitted 22 February, 2001; v1 submitted 26 October, 1999;
originally announced October 1999.
