Dataset Information

Comparing machines and humans on a visual categorization test.

ABSTRACT: Automated scene interpretation has benefited from advances in machine learning, and restricted tasks, such as face detection, have been solved with sufficient accuracy for restricted settings. However, the performance of machines in providing rich semantic descriptions of natural scenes from digital images remains highly limited and hugely inferior to that of humans. Here we quantify this "semantic gap" in a particular setting: We compare the efficiency of human and machine learning in assigning an image to one of two categories determined by the spatial arrangement of constituent parts. The images are not real, but the category-defining rules reflect the compositional structure of real images and the type of "reasoning" that appears to be necessary for semantic parsing. Experiments demonstrate that human subjects grasp the separating principles from a handful of examples, whereas the error rates of computer programs fluctuate wildly and remain far behind that of humans even after exposure to thousands of examples. These observations lend support to current trends in computer vision such as integrating machine learning with parts-based modeling.

SUBMITTER: Fleuret F

PROVIDER: S-EPMC3203755 | biostudies-literature | 2011 Oct

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Comparing machines and humans on a visual categorization test.

Fleuret François F Li Ting T Dubout Charles C Wampler Emma K EK Yantis Steven S Geman Donald D

Proceedings of the National Academy of Sciences of the United States of America 20111017 43

Automated scene interpretation has benefited from advances in machine learning, and restricted tasks, such as face detection, have been solved with sufficient accuracy for restricted settings. However, the performance of machines in providing rich semantic descriptions of natural scenes from digital images remains highly limited and hugely inferior to that of humans. Here we quantify this "semantic gap" in a particular setting: We compare the efficiency of human and machine learning in assigning ...[more]

PMID: 22006295

Similar Datasets

Project description:The hippocampus may play a role in categorization because of the need to differentiate stimulus categories (pattern separation) and to recognize category membership of stimuli from partial information (pattern completion). We hypothesized that the hippocampus would be more crucial for categorization of low-density (few relevant features) stimuli-due to the higher demand on pattern separation and pattern completion-than for categorization of high-density (many relevant features) stimuli. Using a touchscreen apparatus, rats were trained to categorize multiple abstract stimuli into two different categories. Each stimulus was a pentagonal configuration of five visual features; some of the visual features were relevant for defining the category whereas others were irrelevant. Two groups of rats were trained with either a high (dense, n = 8) or low (sparse, n = 8) number of category-relevant features. Upon reaching criterion discrimination (≥75% correct, on 2 consecutive days), bilateral cannulas were implanted in the dorsal hippocampus. The rats were then given either vehicle or muscimol infusions into the hippocampus just prior to various testing sessions. They were tested with: the previously trained stimuli (trained), novel stimuli involving new irrelevant features (novel), stimuli involving relocated features (relocation), and a single relevant feature (singleton). In training, the dense group reached criterion faster than the sparse group, indicating that the sparse task was more difficult than the dense task. In testing, accuracy of both groups was equally high for trained and novel stimuli. However, both groups showed impaired accuracy in the relocation and singleton conditions, with a greater deficit in the sparse group. The testing data indicate that rats encode both the relevant features and the spatial locations of the features. Hippocampal inactivation impaired visual categorization regardless of the density of the category-relevant features for the trained, novel, relocation, and singleton stimuli. Hippocampus-mediated pattern completion and pattern separation mechanisms may be necessary for visual categorization involving overlapping irrelevant features.

Dataset Information

Comparing machines and humans on a visual categorization test.

Publications

Comparing machines and humans on a visual categorization test.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets