Pentland 1992 : This paper is cited in the following contexts: CiteSeer@NUS Home/Search:

Dynamic Model of Visual Recognition Predicts Neural Response.. - Rao, al. (1995) (31 citations) .... the update equations above can be simplified considerably by noting that the basis filters that form the rows of the feedforward matrix ( see learning rule below) also approximately decorrelate their input, thereby effectively diagonalizing the noise covariance matrices (see also [Pentland, 1992] for related ideas) This allows 11 the recursive update rules to be implemented locally in an efficient manner through the use of recurrent axon collaterals. Equation 33 forms the heart of the dynamic state estimation algorithm. Note that Equation 3 which was derived via gradient descent in ....
.... filters have been used in computer vision and robotics for tackling a wide variety of problems ranging from motion estimation to contour tracking [Hallam, 1983; Broida and Chellappa, 1986; Ayache and Faugeras, 1986; Matthies et al. 1989; Blake and Yuille, 1992; Dickmanns and Mysliwetz, 1992; Pentland, 1992] Much of this work crucially hinges on the ability to formulate accurate dynamic physical models of the object properties being estimated. The formulation of such hand coded models however becomes increasingly difficult in more complex dynamic environments. A crucial difference between the ....

Cortical Mechanisms of Visual Recognition and Learning: A.. - Rao (1997)

.... dynamic models, picking values for U and V according to the physics of the dynamic system or other forms of a priori knowledge of the task at hand [Ayache and Faugeras, 1986; Blake and Yuille, 1992; Broida and Chellappa, 1986; Dickmanns and Mysliwetz, 1992; Hallam, 1983; Matthies et al. 1989; Pentland, 1992] However, in complex dynamic environments, the formulation of such hand coded models becomes increasingly difficult. An interesting alternative [Rao and Ballard, 1996a] is to initialize the matrices U and V to small random values, and then adapt these values in response to input data, thereby ....
....U and V , could be learned and refined by the organism during periods of exposure to the visual environment. Similar ideas have been suggested by a number of other authors in a variety of contexts [MacKay, 1956; Grossberg, 1976; Barlow, 1985; Harth et al. 1987; Albus, 1991; Mumford, 1992;
Pentland, 1992; Kawato et al. 1993; Hinton et al. 1995; Dayan et al. 1995; Softky, 1996] 9.1 Neural Network Implementation of the Kalman Filter To see how a Kalman filter like prediction mechanism may be implemented by cortical neurons, first note that the basic operation required by the Kalman filter ....

Dynamic Model of Visual Recognition Predicts Neural Response.. - Rao, Ballard (1997)   (31 citations)

.... the update equations above can be simplified considerably by noting that the basis filters that form the rows of the feedforward matrix W ( U T ; see learning rule below) also approximately decorrelate their input, thereby effectively diagonalizing the noise covariance matrices (see also [Pentland, 1992] for related ideas) This allows the recursive update rules to be implemented locally in an efficient manner through the use of recurrent axon collaterals. Equation 33 forms the heart of the dynamic state estimation algorithm. Note that Equation 3 which was derived via gradient descent in Section ....

Robust Kalman Filters for Prediction, Recognition, and Learning - Rao (1996)  

....process. Most traditional applications of the Kalman filter employ hard wired dynamic models inferred from a priori knowledge of the task at hand [Ayache and Faugeras, 1986; Blake and Yuille, 1992; Broida and Chellappa, 1986; Dickmanns and Mysliwetz, 1992; Hallam, 1983; Matthies et al. 1989; Pentland, 1992] These applications depend crucially on the ability to formulate accurate dynamic physical models of the object properties being estimated. In complex dynamic environments, the formulation of such hand coded models becomes increasingly difficult. An alternate approach is to learn an internal ....

An Optimal Estimation Approach to Visual Perception and Learning - Rao (1999)   (10 citations)

.... dynamic models, picking values for U and V according to the physics of the dynamic system or other forms of a priori knowledge of the task at hand [ Ayache and Faugeras, 1986; Blake and Yuille, 1992; Broida and Chellappa, 1986; Dickmanns and Mysliwetz, 1992; Hallam, 1983; Matthies et al. 1989; Pentland, 1992 ] However, in complex dynamic environments, the formulation of such hand coded models becomes increasingly difficult. A more tractable alternative is to initialize the matrices U and V to small random values, and then adapt these values on line in response to input data, thereby learning an ....
....of exposure to the visual environment. Similar ideas have been suggested by a number of other authors in a variety of contexts [ Albus, 1991; Barlow, 1985; Dayan et al. 1995; Grossberg, 1976; Harth et al. 1987; Hinton et al. 1995; Kawato et al. 1993; MacKay, 1956; Mumford, 1992; Pece, 1992; Pentland, 1992; Softky, 1996 ] Given that the cortex possesses roughly the same neuroanatomical inputoutput structure and pattern of connections across many different cortical areas [ Creutzfeldt, 1977; Barlow, 1985; Pandya et al. 1988 ] a reasonable question to ask is whether a given approach to cortical ....