"Representation is representation of similarities"
Representation is representation of similarities Edelman S., ‘Behavioral and Brain Sciences,’ 1998. Naresh P. Cuntoor Introduction Naresh P. Cuntoor Groundwork Distal shape space Proximal shape space Representing similarity – distinctness, NN preservation, full similarity spectrum Distal to proximal mapping, M - constraints and composition, distance rank preservation Naresh P. Cuntoor Analysis of Mapping M f 4 f 3 f 2 f1 f1(p): Geometry f2(p,z): Imaging f3(p,z): Measurements f4(p): Dimensionality reduction •f4 and f3 need to counteract the z-dependency of f2 •Absolute invariance not necessary –Need: influence of shape space changes > view space changes Naresh P. Cuntoor Representation = Measurement + Dimensionality Reduction •Another example: Sarkar’s face space to affine space Naresh P. Cuntoor Chorus of prototypes An ensemble of tuned classifiers Smooth response degradation Naresh P. Cuntoor Similarity Levels: basic, subordinate, superordinate Features of similarity: pi(A): ith classifier Measures of similarity Naresh P. Cuntoor What’s the brain doing? Novel objects – how chorus deals with it New Pandemonium –feature demons, cognitive demons, master demons ‘Democracy’ in chorus Perception of similarity – ppl. classify maps Naresh P. Cuntoor Experiments and Predictions To test second order isormophism Computer rendered 3D animal shapes and nonsense shapes Predicting distortion in the MDS setting – can chorus do well? Parameter space distances – some more important? Priming – how does it affect? Neurobiology – columns in IT cortex Qualia – attributes of objects Scene richness – humans don’t see everything Bottom-up vs. top-down – Sinha’s STICKS approach Naresh P. Cuntoor Challenges Naresh P. Cuntoor