摘要 :
Conventional paradigms of machine learning assume all the training data are available when learning starts. However, in lifelong learning, the examples are observed sequentially as learning unfolds, and the learner should continua...
展开
Conventional paradigms of machine learning assume all the training data are available when learning starts. However, in lifelong learning, the examples are observed sequentially as learning unfolds, and the learner should continually explore the world and reorganize and refine the internal model or knowledge of the world. This leads to a fundamental challenge: How to balance long-term and short-term goals and how to trade-off between information gain and model complexity. These questions boil down to 'what objective functions can best guide a lifelong learning agent.' Here we develop a sequential Bayesian framework for lifelong learning, build a taxonomy of lifelong-learning paradigms, and examine information-theoretic objective functions for each paradigm, with an emphasis on active learning. The objective functions can provide theoretical criteria for designing algorithms and determining effective strategies for selective sampling, representation discovery, knowledge transfer, and continual update over a lifetime of experience.
收起
摘要 :
We propose a new active learning framework where the expert labeler is allowed to decline to label any example. This may be necessary because the true label is unknown or because the example belongs to a class that is not part of...
展开
We propose a new active learning framework where the expert labeler is allowed to decline to label any example. This may be necessary because the true label is unknown or because the example belongs to a class that is not part of the real training problem. We show that within this framework, popular active learning algorithms (such as Simple) may perform worse than random selection because they make so many queries to the unlabelable class. We present a method by which any active learning algorithm can be modified to avoid unlabelable examples by training a second classifier to distinguish between the labelable and unlabelable classes. We also demonstrate the effectiveness of the method on two benchmark data sets and a real-world problem.
收起
摘要 :
This thesis describes an implemented system called NODDY for acquiring procedures from examples presented by a teacher. Acquiring procedures from examples involves several different generalization tasks. Generalization is an under...
展开
This thesis describes an implemented system called NODDY for acquiring procedures from examples presented by a teacher. Acquiring procedures from examples involves several different generalization tasks. Generalization is an underconstrained task, and the main issue of machine learning is how to deal with this underconstraint. The thesis presents two principles is to exploit domain based constraints. The second principle is to avoid spurious generalizations be requiring justification before adapting a generalization. NODDY demonstrates several different ways of justifying a generalization and proposes a way of ordering and searching a space of candidate generalizations based on how much evidence would be required to justify each generalization. Acquiring procedures also involves three types of constructive generalization: inferring loops ( a kind of group), inferring complex reactions and state variables, and inferring predicates. NODDY demonstrates three constructive generalization for these kinds of generalization.
收起
摘要 :
IES has published a Research Networks report Exploring e-Learning which charts the growing popularity of the media to provide learning in a wide range of subject areas to a wide audience. This preliminary research identified the i...
展开
IES has published a Research Networks report Exploring e-Learning which charts the growing popularity of the media to provide learning in a wide range of subject areas to a wide audience. This preliminary research identified the importance of supporting e-learners. The research also identified that there is a growing use of e-learning to provide not only computer and IT skills, but increasingly, to provide management and other soft skills, which has implications on the type of support needed. e-learning commentators have identified the major issues with regard to supporting learners as: providing appropriate learning environments to enable learners to maximize their learning at what may be their desk. The main issues here are interruptions, time allocation, and the conduciveness of the environment for sustained concentration; creating an electronic learning culture and persuading people that computer-based learning is of equivalent status to more traditional forms of learning, and of equipping people with the skills and confidence to fully benefit from its potential; providing personal support for learners throughout the learning experience and beyond; and overcoming the limitations of e-learning in soft skills development. Unlike many other learning media, electronic learning does not easily provide a full range styles. It has been suggested that it is better suited to those who enjoy theorizing, rather than those for whom practice is important. Similarly it may be less able to help individuals develop certain skill sets that might require insight, support and feedback, or to develop a learning community that depends on mutual trust and the formation of networks. The aim of this project has been to build upon the Institute's work in the field of e-learning by exploring how organizations are supporting their e-learning provision in practice to maximize its efficiency, and how they are supporting e-learners beyond the screen when it comes to the development of advanced skills. In particular the research has focused on the following: initiating and encouraging e-learning; supporting time spent e-learning; maximizing the e-learning experience; and the role of the line manager.
收起
摘要 :
Peace Corps training programs around the world have been teaching hundreds oflanguages to thousands of Volunteers since 1961. Because of its experience with a number of languages (over 200) and the variety of contexts for teaching...
展开
Peace Corps training programs around the world have been teaching hundreds oflanguages to thousands of Volunteers since 1961. Because of its experience with a number of languages (over 200) and the variety of contexts for teaching, Peace Corps has had to find reliable answers to the important questions about the language learning process that are most frequently asked by learners and teachers.
收起
摘要 :
Any successful attempt at explaining and replicating the complexity and generality of human and animal learning will require the integration of a variety of learning mechanisms. Here we introduce a computational model which integr...
展开
Any successful attempt at explaining and replicating the complexity and generality of human and animal learning will require the integration of a variety of learning mechanisms. Here we introduce a computational model which integrates associative learning and reinforcement learning. We contrast the integrated model with associative learning and reinforcement learning models in two simulation studies. The first simulation demonstrates performance advantages for the integrated model in an environment with a dynamic and diverse reward structure. The second simulation contrasts the performances of the three models in a classic latent learning experiment (Blodgett, 1929), demonstrating advantages for the integrated model in predicting and explaining the behavioral data.
收起
摘要 :
The proposed research had the goal of developing computational models of human inferences that bridge the gap between human and machine learning. This research focused on two components of cognition: learning about categories and ...
展开
The proposed research had the goal of developing computational models of human inferences that bridge the gap between human and machine learning. This research focused on two components of cognition: learning about categories and their properties, and learning about causal and social relations. Research was completed successfully in both of these areas, resulting in a new unifying framework for models of category learning, new models for how people form and use high-level generalizations in causal learning, and new methods for predicting people's preferences and the relationships between them. The grant supported a total of 29 publications over three years, including one conference paper that won a best student paper prize, and provided support for five graduate students.
收起
摘要 :
Reinforcement learning with reward shaping is a well-established but often computationally expensive approach to multiagent problems. Agent partitioning can assist in this computational complexity by treating each partition of age...
展开
Reinforcement learning with reward shaping is a well-established but often computationally expensive approach to multiagent problems. Agent partitioning can assist in this computational complexity by treating each partition of agents as an independent problem. We introduce a novel agent partitioning approach called Reward/Utility-Based Impact (RUBI). RUBI finds an effective partitioning of agents while requiring no prior domain knowledge, provides better performance by discovering a non-trivial agent partitioning, and leads to faster simulations. We test RUBI in the Air Tra c Flow Management Problem, where there are simultaneously tens of thousands of aircraft affecting the system and no intuitive similarity metric between agents. When partitioning with RUBI in the ATFMP, there is a 37% increase in performance, with a 510x speed up per simulation step over non-partitioning approaches.
收起
