Download
1 / 59
590 likes | 746 Views
Computational Architectures in Biological Vision, USC. Lecture 12. Visual Attention Reading Assignments: None. Several Forms of Attention. Attention and eye movements: - overt attention (with eye movements) - covert attention (without eye movements) Bottom-up and top-down control:
E N D
Computational Architectures in Biological Vision, USC • Lecture 12. Visual Attention • Reading Assignments: • None
Several Forms of Attention • Attention and eye movements: • - overt attention (with eye movements) • - covert attention (without eye movements) • Bottom-up and top-down control: • - bottom-up control • based on image features • very fast (up to 20 shifts/s) • involuntary / automatic • - top-down control • may target inconspicuous locations in visual scene • slower (5 shifts/s or fewer; like eye movements) • volitional • Control and modulation: • - direct attention towards specific visual locations • - attention modulates early visual processing at attended location
What is attention then? • Attention is often described as an information processing bottleneck. • Controls access to higher levels of processing, short-term memory and consciousness. • Hence, the strategy nature has developed to cope with information overload is to break down the problem of analyzing a visual scene: • from a massively parallel approach • to a rapid sequence of circumscribed recognitions.
First Computational Model • Koch & Ullman, • Hum. Neurobiol., 1895 • Introduce concept • of a single topo- • graphic saliency • map. • Most salient • location selected • by a winner-take-all • network.
Shifter Circuits • Anderson & van Essen, PNAS, 1987 • Information dynamically routed through • cortical hierarchy. Yields rotation- and • scale-independent representation.
Shifter Circuits (cont.) • Olshausen et al., J Neurosci, 1993 • Implemented shifter circuits and demonstrated proof of concept. • Control neurons in the pulvinar send the (attention-based) control signals that will determine the “passing” region of the circuit, through a modulation of intracortical connection weights. • Perform recognition • using associative • memory at top • level.
Selective Tuning Model Tsotsos et al., Artificial Intelligence, 1995 - attention modulates neurons to earliest levels; wherever there is a many-to-one mapping - signal interference controlled by surround inhibition throughout processing network • task knowledge biases computations throughout processing network - attentional control is local, distributed and internal - competition is based on WTA (different form than previous models) - pyramid representation with reciprocal convergence and divergence neuron ‘sees’ this receptive field subject ‘attends’ to single item
The basic idea (BBS 1990)
Selective Tuning Model unit of interest at top processing pyramid pass input pathways inhibited pathways Kastner, De Weerd, Desimone, Ungerleider, 1998 Caputo & Guerra 1998 Bahcall & Kowler 1999 Vanduffel, Tootell, Orban 2000 Smith et al. 2000
Guided Search • Wolfe, Psychonomic Bull. & Rev., 1994 • How can we combine information from several modalities? Use top-down (task-dependent) weighting.
