Related Experiment Video
Updated: May 2, 2026
13:40
Combining Computer Game-Based Behavioural Experiments With High-Density EEG and Infrared Gaze Tracking
Published on: December 16, 2010
16.7K
Do we really need a large number of visual prompts?
Youngeun Kim1, Yuhang Li1, Abhishek Moitra1
1Department of Electrical Engineering, Yale University, New Haven, CT, USA.
Summary
Visual Prompt Tuning (VPT) can be made more efficient. This study introduces Prompt Condensation (PC) to reduce prompts by 70% without losing accuracy in vision transformers.
Area of Science:
- Computer Vision
- Machine Learning
- Artificial Intelligence
Background:
- Parameter-efficient transfer learning is crucial for adapting models on resource-constrained edge devices.
- Visual Prompt Tuning (VPT) offers competitive performance by prepending learnable prompts but increases computational overhead due to more input tokens.
Purpose of the Study:
- To analyze the impact of prompt quantity on vision transformer fine-tuning performance and self-attention mechanisms.
- To develop a method for reducing prompt usage while maintaining accuracy.
Main Methods:
- Theoretical and empirical analysis of prompt number effects on vision transformer performance.
- Proposal of a Prompt Condensation (PC) technique to mitigate performance loss with fewer prompts.
Main Results:
- Increasing prompts does not yield linearly improved fine-tuning performance.
- The proposed Prompt Condensation (PC) technique effectively reduces the number of prompts by approximately 70% on FGVC and VTAB-1k tasks.
- Accuracy is maintained despite the significant reduction in prompt usage.
Conclusions:
- Optimizing prompt quantity is essential for efficient parameter-efficient transfer learning.
- Prompt Condensation (PC) offers a viable solution for reducing computational overhead in Visual Prompt Tuning (VPT) without compromising model accuracy.
Related Concept Videos
01:24Vision
48.6K
Vision is the result of light being detected and transduced into neural signals by the retina of the eye. This information is then further analyzed and interpreted by the brain. First, light enters the front of the eye and is focused by the cornea and lens onto the retina—a thin sheet of neural tissue lining the back of the eye. Because of refraction through the convex lens of the eye, images are projected onto the retina upside-down and reversed.
48.6K
01:17Accessory Structures of the Eye
4.3K
Optical perception, or vision, is an extraordinary sense dependent on converting light signals received via the ocular organs. These organs, known as eyes, are securely positioned within the bony cavities of the skull, called orbits. The orbits serve a dual purpose: a protective shield for the ocular globes and a stable attachment point for the soft ocular tissues. The eye's external protective mechanisms include the eyelids, which are edged with lashes that act as a barrier against foreign...
4.3K
01:22Photoreceptors and Visual Pathways
8.5K
At the molecular level, visual signals trigger transformations in photopigment molecules, resulting in changes in the photoreceptor cell's membrane potential. The photon's energy level is denoted by its wavelength, with each specific wavelength of visible light associated with a distinct color. The spectral range of visible light, classified as electromagnetic radiation, spans from 380 to 720 nm. Electromagnetic radiation wavelengths exceeding 720 nm fall under the infrared category,...
8.5K
01:15Depth Perception and Spatial Vision
2.7K
Depth perception is the ability to perceive objects three-dimensionally. It relies on two types of cues: binocular and monocular. Binocular cues depend on the combination of images from both eyes and how the eyes work together. Since the eyes are in slightly different positions, each eye captures a slightly different image. This disparity between images, known as binocular disparity, helps the brain interpret depth. When the brain compares these images, it determines the distance to an object.
2.7K
01:26Visual System
2.3K
Light enters the eye through the cornea, a transparent, dome-shaped surface covering the surface of the eyeball that helps to direct and focus incoming light. This light is then channeled toward the pupil, an adjustable opening whose size is controlled by the iris. The iris, a pigmented muscle, regulates the amount of light entering the eye by contracting or dilating the pupil, thereby ensuring optimal light levels for clear vision.
Once through the pupil, the light passes through the lens, a...
Once through the pupil, the light passes through the lens, a...
2.3K
01:29Non-Verbal Cues
774
Non-verbal communication extends beyond gestures and facial expressions to include vocal elements known as paralanguage. Paralanguage consists of non-verbal vocal cues such as pitch, loudness, speech rate, pauses, and non-verbal vocalizations like laughter, sighs, and moans. These elements not only accompany speech but also provide critical emotional and contextual information.The Role of Paralanguage in CommunicationParalanguage adds depth to spoken language by conveying emotions and...
774