When computer vision works more like a brain, it sees more like people do
MIT researchers developed a computer vision model mimicking human brain processing, trained on monkey IT cortex data, enhancing object recognition and resistance to adversarial attacks, promoting collaboration between neuroscience and AI.
Read original article
Researchers at MIT have developed a computer vision model that mimics the brain's processing of visual information, specifically targeting the inferior temporal (IT) cortex, which is crucial for object recognition in humans and primates. Led by Professor James DiCarlo, the team trained an artificial neural network using data from the IT cortex of monkeys, resulting in a model that not only improved its ability to identify objects but also aligned more closely with human visual perception. This neurally aligned model demonstrated enhanced robustness against adversarial attacks—small distortions in images designed to confuse AI systems—indicating a more human-like processing capability. The findings suggest that integrating biological principles into AI development can yield significant advancements in computer vision, benefiting both fields by providing insights into human vision mechanisms and improving AI robustness. The research highlights the potential for a collaborative exchange between neuroscience and artificial intelligence, fostering progress in understanding and replicating human-like vision in machines.
- MIT researchers have created a computer vision model that mimics human brain processing.
- The model was trained using neural data from the monkey IT cortex, enhancing object recognition.
- It showed improved resistance to adversarial attacks compared to standard models.
- The study emphasizes the benefits of integrating neuroscience insights into AI development.
- This research fosters collaboration between neuroscience and artificial intelligence fields.
Related
Mind-reading AI recreates what you're looking at with accuracy
Artificial intelligence excels in reconstructing images from brain activity, especially when focusing on specific regions. Umut Güçlü praises the precision of these reconstructions, enhancing neuroscience and technology applications significantly.
MIT researchers advance automated interpretability in AI models
MIT researchers developed MAIA, an automated system enhancing AI model interpretability, particularly in vision systems. It generates hypotheses, conducts experiments, and identifies biases, improving understanding and safety in AI applications.
NeuroAI paper proposes "Embodied Turing Test" to evaluate AI (2023)
Neuroscience is vital for AI progress, with a focus on NeuroAI. The embodied Turing test assesses AI's sensorimotor skills, highlighting the need for research into biological intelligence to enhance AI capabilities.
Sapiens: Foundation for Human Vision Models
The "Sapiens" models enhance human-centric vision tasks through self-supervised pretraining on 300 million images, showing strong generalization and scalability, outperforming benchmarks in several datasets.
Novel Chinese computing architecture 'inspired by human brain' can lead to AGI
Scientists in China have developed a brain-inspired computing architecture that focuses on internal complexity, potentially leading to artificial general intelligence (AGI) and more efficient AI systems.
2 commentsUnlike the human vision system, there's no mention of the fovea, the foveated gaze which forces us to spend attention and build a mental model of the world, nor the need to focus on objects at various distances. While it's astounding how well deep networks have handled image input, it's definitely not like that of a human, or most other animals.
Related
Mind-reading AI recreates what you're looking at with accuracy
Artificial intelligence excels in reconstructing images from brain activity, especially when focusing on specific regions. Umut Güçlü praises the precision of these reconstructions, enhancing neuroscience and technology applications significantly.
MIT researchers advance automated interpretability in AI models
MIT researchers developed MAIA, an automated system enhancing AI model interpretability, particularly in vision systems. It generates hypotheses, conducts experiments, and identifies biases, improving understanding and safety in AI applications.
NeuroAI paper proposes "Embodied Turing Test" to evaluate AI (2023)
Neuroscience is vital for AI progress, with a focus on NeuroAI. The embodied Turing test assesses AI's sensorimotor skills, highlighting the need for research into biological intelligence to enhance AI capabilities.
Sapiens: Foundation for Human Vision Models
The "Sapiens" models enhance human-centric vision tasks through self-supervised pretraining on 300 million images, showing strong generalization and scalability, outperforming benchmarks in several datasets.
Novel Chinese computing architecture 'inspired by human brain' can lead to AGI
Scientists in China have developed a brain-inspired computing architecture that focuses on internal complexity, potentially leading to artificial general intelligence (AGI) and more efficient AI systems.