Abstract A linear mixture of experts is used to combine three standard IR systems. The parameters... more Abstract A linear mixture of experts is used to combine three standard IR systems. The parameters for the mixture are determined automatically through training on document relevance assessments via optimization of a rank-order statistic which is empirically correlated with average precision. The mixture improves performance in some cases and degrades it in others, with the degradations possibly due to training techniques, model strength, and poor performance of the individual experts.
Abstract Savage [1] proposed analyzing active sampling problems as decision problems in which the... more Abstract Savage [1] proposed analyzing active sampling problems as decision problems in which the goal is to maximize expected utility, relative to a probability distribution describing one's beliefs. In the past 20 years this framework has been applied to several psychological tasks [2]. We use this framework to model eye movements in a concept formation task [3],[4].
Representation and Induction of Finite State Machines using Time-Delay Neural Networks Daniel S. ... more Representation and Induction of Finite State Machines using Time-Delay Neural Networks Daniel S. Clouse Computer Science & Engineering Dept. University of California, San Diego La Jolla, CA 92093-0114 dclouse@ucsd.edu C. Lee Giles NEC Research Institute 4 Independence Way Princeton, NJ 08540 giles@research.nj .nec.com Bill G. Horne NEC Research Institute 4 Independence Way Princeton, NJ 08540 horne@ research. nj. nec. corn Garrison W. Cottrell Computer Science & Engineering Dept.
Abstract Deciding which piece of information to acquire or attend to is fundamental to perception... more Abstract Deciding which piece of information to acquire or attend to is fundamental to perception, categorization, medical diagnosis, and scientific inference. Four statistical theories of the value of information—information gain, Kullback-Liebler distance, probability gain (error minimization), and impact—are equally consistent with extant data on human information acquisition.
Brightness and color constancy is a fundamental problem faced in computer vision and by our own v... more Brightness and color constancy is a fundamental problem faced in computer vision and by our own visual system. We easily recognize objects despite changes in illumination, but without a mechanism to cope with this, many object and face recognition systems perform poorly. In this paper we compare approaches in computer vision and computational neuroscience for inducing brightness and color constancy based on their ability to improve recognition.
Abstract In previous work, we modeled Mondloch's behavioral data [Mondloch et al., 2002] on adult... more Abstract In previous work, we modeled Mondloch's behavioral data [Mondloch et al., 2002] on adult face discrimination [Zhang and Cottrell, 2004]. We found that our standard model [Dailey and Cottrell, 1999, Dailey et al., 2002] was overly holistic and that by adding local feature processing we could qualitatively match the human data. However, further investigation has lead us to reconsider our conclusions.
To help elucidate the ideas behind our model we provide a demonstration video, NIMBLE_Demo. mp4, ... more To help elucidate the ideas behind our model we provide a demonstration video, NIMBLE_Demo. mp4, with the supplementary materials. The demo is an animated version of the model discussed in the paper applied to a subset of the AR dataset [8]. It first extracts ICA features and acquires fixations/samples from 10 images, each representing a different individual. This is followed by PCA and evaluation of performance using four test images per class.
Abstract Plunkett & Marchman (1993) showed that a neural network trained on an incrementally expa... more Abstract Plunkett & Marchman (1993) showed that a neural network trained on an incrementally expanded training set was able to master the past tense and show the U-shaped learning pattern characteristic of children. In Jackson, Constandse & Cottrell (1996) we argued that Plunkett & Marchman's restriction of the training set was unrealistic and proposed a model of selective attention that enabled our network to master the past tense without external restrictions on its training set.
Abstract This paper describes the organization and use of an interactive computer program called ... more Abstract This paper describes the organization and use of an interactive computer program called ISCON, which aids in the development of connectionist models of cognitive processes and neural systems.
Abstract Connectionist attractor networks have played a central role in many cognitive models inv... more Abstract Connectionist attractor networks have played a central role in many cognitive models involving associative memory and soft constraint satisfaction. While early attractor networks used step activation functions, permitting the construction of attractors for only binary (or bipolar) patterns, much recent work has focused on networks with continuous sigmoidal activation functions. The incorporation of sigmoidal processing elements allows for the use of expressive real vector representations in attractor networks.
Neighborhood and Position Effects Interact in Naming Latency Jeanne C. Milostan (JMILOSTA@ CS. UC... more Neighborhood and Position Effects Interact in Naming Latency Jeanne C. Milostan (JMILOSTA@ CS. UCSD. EDU) Victor Ferreira (FERREIRA@ PSY. UCSD. EDU) Garrison W. Cottrell (GARY@ CS. UCSD.
Abstract We present the results from a large set of connectionist simulations exploring the effec... more Abstract We present the results from a large set of connectionist simulations exploring the effect of subject omission (prodrop) on languages with an SVO word order. We show that pro-drop only affects the learnability of the languages if there are no cues available to tell nouns and verbs apart. We also argue that neither creoles nor Mandarin Chinese instantiate the language type which was unlearnable in the simulations.
Abstract What is the difference between processing faces and other objects such as letters? What ... more Abstract What is the difference between processing faces and other objects such as letters? What makes humans face experts, and what makes this expertise different from other identification skills? It is well known that people are very sensitive to the configural information in faces. How does the sensitivity to face configuration compare to sensitivity to configurations of other stimuli? To investigate these issues, Nishimura et al.(2004) designed a test to contrast two types of processing using the same stimuli.
Abstract The ability to decide between multiple fixation targets in complex visual environments i... more Abstract The ability to decide between multiple fixation targets in complex visual environments is essential for our survival. Evolution has refined this process to be both rapid and cheap, allowing us to perform over 100,000 saccades a day. Previous models for visual decision making have focused on maximizing reward magnitude or expected value (EV= probability of reward× magnitude of reward). However, such methods fail to incorporate utility, or happiness derived from reward, optimizing strictly on nominal reward values.
Abstract Left-right asymmetries have been noted in tasks requiring the classification of many vis... more Abstract Left-right asymmetries have been noted in tasks requiring the classification of many visual stimuli, including Navon figures, spatial frequency gratings, and faces. The Double Filtering by Frequency (DFF) model (Ivry & Robertson, 1998), which postulates asymmetric frequency filtering on task-relevant frequency bands, has been implemented to account for asymmetric processing of each stimulus type above, but does not provide a fully mechanistic explanation, nor does it have direct neural correlates.
Abstract The performance of a connectionist network, in which some resources are absent or damage... more Abstract The performance of a connectionist network, in which some resources are absent or damaged is examined as a function of various learning parameters. A learning environment is created by generating a set of random “prototypes” and clusters of exemplar vectors surrounding each prototype. An autoencoder is trained on the patterns. The robustness of each learned item is measured as a function of the time at which it was “acquired” by the network and its overall frequency in the environment.
Humans improve their performance by means of a variety of learning strategies, including both gra... more Humans improve their performance by means of a variety of learning strategies, including both gradual statistical induction from experience and rapid incorporation of advice. In many learning environments, these strategies may interact in complementary ways. The focus of this work is on cognitively plausible models of multistrategy learning involving the integration of inductive generalization and learning\ by being told".
Abstract In contrast to the wealth of saliency models in the vision literature, there is a relati... more Abstract In contrast to the wealth of saliency models in the vision literature, there is a relative paucity of models exploring auditory saliency. In this work, we integrate the approaches of (Kayser, Petkov, Lippert, & Logothetis, 2005) and (Zhang, Tong, Marks, Shan, & Cottrell, 2008) and propose a model of auditory saliency. The model combines the statistics of natural soundscapes and the recent past of the input signal to predict the saliency of an auditory stimulus in the frequency domain.
Abstract We present a neurocomputational model of visual object processing, which takes photograp... more Abstract We present a neurocomputational model of visual object processing, which takes photographic inputs and creates topographic stimulus representations on the hidden layer. We perform multi-voxel pattern analysis on the activations of hidden units and simulate contradictory findings from Haxby et al.(2001) and Spiridon and Kanwisher (2002) within a single model.
Abstract A linear mixture of experts is used to combine three standard IR systems. The parameters... more Abstract A linear mixture of experts is used to combine three standard IR systems. The parameters for the mixture are determined automatically through training on document relevance assessments via optimization of a rank-order statistic which is empirically correlated with average precision. The mixture improves performance in some cases and degrades it in others, with the degradations possibly due to training techniques, model strength, and poor performance of the individual experts.
Abstract Savage [1] proposed analyzing active sampling problems as decision problems in which the... more Abstract Savage [1] proposed analyzing active sampling problems as decision problems in which the goal is to maximize expected utility, relative to a probability distribution describing one's beliefs. In the past 20 years this framework has been applied to several psychological tasks [2]. We use this framework to model eye movements in a concept formation task [3],[4].
Representation and Induction of Finite State Machines using Time-Delay Neural Networks Daniel S. ... more Representation and Induction of Finite State Machines using Time-Delay Neural Networks Daniel S. Clouse Computer Science & Engineering Dept. University of California, San Diego La Jolla, CA 92093-0114 dclouse@ucsd.edu C. Lee Giles NEC Research Institute 4 Independence Way Princeton, NJ 08540 giles@research.nj .nec.com Bill G. Horne NEC Research Institute 4 Independence Way Princeton, NJ 08540 horne@ research. nj. nec. corn Garrison W. Cottrell Computer Science & Engineering Dept.
Abstract Deciding which piece of information to acquire or attend to is fundamental to perception... more Abstract Deciding which piece of information to acquire or attend to is fundamental to perception, categorization, medical diagnosis, and scientific inference. Four statistical theories of the value of information—information gain, Kullback-Liebler distance, probability gain (error minimization), and impact—are equally consistent with extant data on human information acquisition.
Brightness and color constancy is a fundamental problem faced in computer vision and by our own v... more Brightness and color constancy is a fundamental problem faced in computer vision and by our own visual system. We easily recognize objects despite changes in illumination, but without a mechanism to cope with this, many object and face recognition systems perform poorly. In this paper we compare approaches in computer vision and computational neuroscience for inducing brightness and color constancy based on their ability to improve recognition.
Abstract In previous work, we modeled Mondloch's behavioral data [Mondloch et al., 2002] on adult... more Abstract In previous work, we modeled Mondloch's behavioral data [Mondloch et al., 2002] on adult face discrimination [Zhang and Cottrell, 2004]. We found that our standard model [Dailey and Cottrell, 1999, Dailey et al., 2002] was overly holistic and that by adding local feature processing we could qualitatively match the human data. However, further investigation has lead us to reconsider our conclusions.
To help elucidate the ideas behind our model we provide a demonstration video, NIMBLE_Demo. mp4, ... more To help elucidate the ideas behind our model we provide a demonstration video, NIMBLE_Demo. mp4, with the supplementary materials. The demo is an animated version of the model discussed in the paper applied to a subset of the AR dataset [8]. It first extracts ICA features and acquires fixations/samples from 10 images, each representing a different individual. This is followed by PCA and evaluation of performance using four test images per class.
Abstract Plunkett & Marchman (1993) showed that a neural network trained on an incrementally expa... more Abstract Plunkett & Marchman (1993) showed that a neural network trained on an incrementally expanded training set was able to master the past tense and show the U-shaped learning pattern characteristic of children. In Jackson, Constandse & Cottrell (1996) we argued that Plunkett & Marchman's restriction of the training set was unrealistic and proposed a model of selective attention that enabled our network to master the past tense without external restrictions on its training set.
Abstract This paper describes the organization and use of an interactive computer program called ... more Abstract This paper describes the organization and use of an interactive computer program called ISCON, which aids in the development of connectionist models of cognitive processes and neural systems.
Abstract Connectionist attractor networks have played a central role in many cognitive models inv... more Abstract Connectionist attractor networks have played a central role in many cognitive models involving associative memory and soft constraint satisfaction. While early attractor networks used step activation functions, permitting the construction of attractors for only binary (or bipolar) patterns, much recent work has focused on networks with continuous sigmoidal activation functions. The incorporation of sigmoidal processing elements allows for the use of expressive real vector representations in attractor networks.
Neighborhood and Position Effects Interact in Naming Latency Jeanne C. Milostan (JMILOSTA@ CS. UC... more Neighborhood and Position Effects Interact in Naming Latency Jeanne C. Milostan (JMILOSTA@ CS. UCSD. EDU) Victor Ferreira (FERREIRA@ PSY. UCSD. EDU) Garrison W. Cottrell (GARY@ CS. UCSD.
Abstract We present the results from a large set of connectionist simulations exploring the effec... more Abstract We present the results from a large set of connectionist simulations exploring the effect of subject omission (prodrop) on languages with an SVO word order. We show that pro-drop only affects the learnability of the languages if there are no cues available to tell nouns and verbs apart. We also argue that neither creoles nor Mandarin Chinese instantiate the language type which was unlearnable in the simulations.
Abstract What is the difference between processing faces and other objects such as letters? What ... more Abstract What is the difference between processing faces and other objects such as letters? What makes humans face experts, and what makes this expertise different from other identification skills? It is well known that people are very sensitive to the configural information in faces. How does the sensitivity to face configuration compare to sensitivity to configurations of other stimuli? To investigate these issues, Nishimura et al.(2004) designed a test to contrast two types of processing using the same stimuli.
Abstract The ability to decide between multiple fixation targets in complex visual environments i... more Abstract The ability to decide between multiple fixation targets in complex visual environments is essential for our survival. Evolution has refined this process to be both rapid and cheap, allowing us to perform over 100,000 saccades a day. Previous models for visual decision making have focused on maximizing reward magnitude or expected value (EV= probability of reward× magnitude of reward). However, such methods fail to incorporate utility, or happiness derived from reward, optimizing strictly on nominal reward values.
Abstract Left-right asymmetries have been noted in tasks requiring the classification of many vis... more Abstract Left-right asymmetries have been noted in tasks requiring the classification of many visual stimuli, including Navon figures, spatial frequency gratings, and faces. The Double Filtering by Frequency (DFF) model (Ivry & Robertson, 1998), which postulates asymmetric frequency filtering on task-relevant frequency bands, has been implemented to account for asymmetric processing of each stimulus type above, but does not provide a fully mechanistic explanation, nor does it have direct neural correlates.
Abstract The performance of a connectionist network, in which some resources are absent or damage... more Abstract The performance of a connectionist network, in which some resources are absent or damaged is examined as a function of various learning parameters. A learning environment is created by generating a set of random “prototypes” and clusters of exemplar vectors surrounding each prototype. An autoencoder is trained on the patterns. The robustness of each learned item is measured as a function of the time at which it was “acquired” by the network and its overall frequency in the environment.
Humans improve their performance by means of a variety of learning strategies, including both gra... more Humans improve their performance by means of a variety of learning strategies, including both gradual statistical induction from experience and rapid incorporation of advice. In many learning environments, these strategies may interact in complementary ways. The focus of this work is on cognitively plausible models of multistrategy learning involving the integration of inductive generalization and learning\ by being told".
Abstract In contrast to the wealth of saliency models in the vision literature, there is a relati... more Abstract In contrast to the wealth of saliency models in the vision literature, there is a relative paucity of models exploring auditory saliency. In this work, we integrate the approaches of (Kayser, Petkov, Lippert, & Logothetis, 2005) and (Zhang, Tong, Marks, Shan, & Cottrell, 2008) and propose a model of auditory saliency. The model combines the statistics of natural soundscapes and the recent past of the input signal to predict the saliency of an auditory stimulus in the frequency domain.
Abstract We present a neurocomputational model of visual object processing, which takes photograp... more Abstract We present a neurocomputational model of visual object processing, which takes photographic inputs and creates topographic stimulus representations on the hidden layer. We perform multi-voxel pattern analysis on the activations of hidden units and simulate contradictory findings from Haxby et al.(2001) and Spiridon and Kanwisher (2002) within a single model.
Uploads
Papers by Garrison W. Cottrell