-
SoftAdapt: Techniques for Adaptive Loss Weighting of Neural Networks with Multi-Part Loss Functions
Authors:
A. Ali Heydari,
Craig A. Thompson,
Asif Mehmood
Abstract:
Adaptive loss function formulation is an active area of research and has gained a great deal of popularity in recent years, following the success of deep learning. However, existing frameworks of adaptive loss functions often suffer from slow convergence and poor choice of weights for the loss components. Traditionally, the elements of a multi-part loss function are weighted equally or their weigh…
▽ More
Adaptive loss function formulation is an active area of research and has gained a great deal of popularity in recent years, following the success of deep learning. However, existing frameworks of adaptive loss functions often suffer from slow convergence and poor choice of weights for the loss components. Traditionally, the elements of a multi-part loss function are weighted equally or their weights are determined through heuristic approaches that yield near-optimal (or sub-optimal) results. To address this problem, we propose a family of methods, called SoftAdapt, that dynamically change function weights for multi-part loss functions based on live performance statistics of the component losses. SoftAdapt is mathematically intuitive, computationally efficient and straightforward to implement. In this paper, we present the mathematical formulation and pseudocode for SoftAdapt, along with results from applying our methods to image reconstruction (Sparse Autoencoders) and synthetic data generation (Introspective Variational Autoencoders).
△ Less
Submitted 27 December, 2019;
originally announced December 2019.
-
A Distributed IP-Based Telecommunication System using SIP
Authors:
Carlton A. Thompson,
Haniph A. Latchman,
Nathan Angelacos,
Bharath Kumar Pareek
Abstract:
Voice over Internet Protocol (VoIP) technologies are integral to modern telecommunications because of their advanced features, flexibility, and economic benefits. Internet Service Providers initially promoted these technologies by providing low cost local and international calling. At present, there is also a great deal of interest in using IP-based technologies to replace traditional small and la…
▽ More
Voice over Internet Protocol (VoIP) technologies are integral to modern telecommunications because of their advanced features, flexibility, and economic benefits. Internet Service Providers initially promoted these technologies by providing low cost local and international calling. At present, there is also a great deal of interest in using IP-based technologies to replace traditional small and large office telephone systems that use traditional PBXs (Private Branch eXchange). Unfortunately, the large majority of the emerging VoIP based office telephone systems have followed the centralized design of traditional public and private telephone systems in which all the intelligence in the system is at the core, with quite expensive hardware and software components and appropriate redundancy for adequate levels of reliability. In this paper, it is argued that a centralized model for an IP-based telecommunications system fails to exploit the full capabilities of Internet-inspired communications and that, very simple, inexpensive, elegant and flexible solutions are possible by deliberately avoiding the centralized approach. This paper describes the design, philosophy and implementation of a prototype for a fully distributed IP-based Telecommunication System (IPTS) that provides the essential feature set for office and home telecommunications, including IP-based long-distance and local calling, and with the support for video as well as data and text. The prototype system was implemented with an Internet-inspired distributed design using open source software, with appropriate customizations and configurations.
△ Less
Submitted 9 December, 2013;
originally announced December 2013.
-
An Exploratory Study of Forces and Frictions affecting Large-Scale Model-Driven Development
Authors:
Adrian Kuhn,
Gail C. Murphy,
C. Albert Thompson
Abstract:
In this paper, we investigate model-driven engineering, reporting on an exploratory case-study conducted at a large automotive company. The study consisted of interviews with 20 engineers and managers working in different roles. We found that, in the context of a large organization, contextual forces dominate the cognitive issues of using model-driven technology. The four forces we identified that…
▽ More
In this paper, we investigate model-driven engineering, reporting on an exploratory case-study conducted at a large automotive company. The study consisted of interviews with 20 engineers and managers working in different roles. We found that, in the context of a large organization, contextual forces dominate the cognitive issues of using model-driven technology. The four forces we identified that are likely independent of the particular abstractions chosen as the basis of software development are the need for diffing in software product lines, the needs for problem-specific languages and types, the need for live modeling in exploratory activities, and the need for point-to-point traceability between artifacts. We also identified triggers of accidental complexity, which we refer to as points of friction introduced by languages and tools. Examples of the friction points identified are insufficient support for model diffing, point-to-point traceability, and model changes at runtime.
△ Less
Submitted 3 July, 2012;
originally announced July 2012.
-
A Personalized System for Conversational Recommendations
Authors:
M. H. Goker,
P. Langley,
C. A. Thompson
Abstract:
Searching for and making decisions about information is becoming increasingly difficult as the amount of information and number of choices increases. Recommendation systems help users find items of interest of a particular type, such as movies or restaurants, but are still somewhat awkward to use. Our solution is to take advantage of the complementary strengths of personalized recommendation sys…
▽ More
Searching for and making decisions about information is becoming increasingly difficult as the amount of information and number of choices increases. Recommendation systems help users find items of interest of a particular type, such as movies or restaurants, but are still somewhat awkward to use. Our solution is to take advantage of the complementary strengths of personalized recommendation systems and dialogue systems, creating personalized aides. We present a system -- the Adaptive Place Advisor -- that treats item selection as an interactive, conversational process, with the program inquiring about item attributes and the user responding. Individual, long-term user preferences are unobtrusively obtained in the course of normal recommendation dialogues and used to direct future conversations with the same user. We present a novel user model that influences both item search and the questions asked during a conversation. We demonstrate the effectiveness of our system in significantly reducing the time and number of interactions required to find a satisfactory item, as compared to a control group of users interacting with a non-adaptive version of the system.
△ Less
Submitted 30 June, 2011;
originally announced July 2011.
