Video Experimental Relacionado
Updated: Sep 9, 2025

04:23
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
2.0K
FRANCIA: Optimización conjunta de tokens y poda de canales estructurales para la inferencia ViT adaptativa
IEEE transactions on pattern analysis and machine intelligence
|September 2, 2025
Resumen
PRANCE acelera los transformadores de visión (ViT) mediante la optimización conjunta de canales y tokens por muestra. Este marco reduce la complejidad computacional y el tamaño del modelo sin sacrificar la precisión, lo que permite una implementación eficiente de ViT.
Área de la Ciencia:
- Visión por computadora
- Inteligencia artificial
- Aprendizaje automático
Sus antecedentes:
- Los Vision Transformers (ViTs) enfrentan desafíos de implementación debido al gran tamaño del modelo y la complejidad cuadrática con el número de tokens.
- Los métodos existentes para acelerar ViTs, como la poda y la reducción de tokens, a menudo usan proporciones fijas y descuidan la optimización conjunta, lo que lleva a una pérdida de precisión.
Objetivo del estudio:
- Introducir PRANCE, un nuevo marco para optimizar conjuntamente los canales activados y los tokens por muestra para acelerar la inferencia ViT.
- Para abordar los desafíos de la computación dinámica del canal y el vasto espacio de decisión en la optimización conjunta.
Principales métodos:
- Desarrolló una meta-red con reparto de peso para el soporte dinámico de canales en capas de autoatención de múltiples cabezas (MHSA) y percepción de múltiples capas (MLP).
- Empleado Proximal Policy Optimization (PPO) a través de un selector ligero para administrar de manera eficiente el problema de optimización combinatoria.
- Se introdujo un mecanismo de capacitación "Resultado para ir" que modela la inferencia ViT como un proceso de decisión de Markov para reducir el espacio de acción y el retraso de recompensa.
Principales resultados:
- Se ha conseguido una reducción de aproximadamente el 50% en las FLOP (operaciones de punto flotante).
- Retenido sólo alrededor del 10% de los tokens de entrada.
- Mantiene la precisión Top-1 sin pérdidas, demostrando ganancias de eficiencia significativas.
Conclusiones:
- PRANCE ofrece un enfoque unificado para acelerar las TI mediante la optimización simultánea de la arquitectura y los datos.
- El marco aborda efectivamente la compensación entre la compresión y la precisión, lo que permite un despliegue eficiente de ViT.
Videos de Conceptos Relacionados
Transformers with Off-Nominal Turns Ratios
205
In scenarios involving parallel transformers with disparate ratings, developing per-unit models requires accommodating off-nominal turns ratios. This situation arises when the selected base voltages are not proportional to the transformer’s voltage ratings. Consider a transformer where the rated voltages are related by the term a. If the chosen voltage bases satisfy a relationship involving term b, term c is defined as the ratio of these bases. This ratio is then substituted into the...
205
Vision
55.3K
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.
55.3K
Transformers in Distribution System
156
Transformers in distribution systems can be broadly categorized into distribution substation transformers and other distribution transformers. They are crucial for stepping down high transmission voltages to levels suitable for distribution and end-user applications.
Distribution substation transformers come in various ratings and typically use mineral oil for insulation and cooling. To prevent moisture and air from entering the oil, some transformers use an inert gas like nitrogen to fill the...
Distribution substation transformers come in various ratings and typically use mineral oil for insulation and cooling. To prevent moisture and air from entering the oil, some transformers use an inert gas like nitrogen to fill the...
156
Improving Translational Accuracy
11.8K
Base complementarity between the three base pairs of mRNA codon and the tRNA anticodon is not a failsafe mechanism. Inaccuracies can range from a single mismatch to no correct base pairing at all. The free energy difference between the correct and nearly correct base pairs can be as small as 3 kcal/ mol. With complementarity being the only proofreading step, the estimated error frequency would be one wrong amino acid in every 100 amino acids incorporated. However, error frequencies observed in...
11.8K
Reducing Line Loss
193
In a three-phase circuit, line loss is an indicator of energy dissipated as heat due to the resistance of transmission lines. To address this, incorporating transformers into the system—a step-up transformer at the source and a step-down transformer at the load—is a strategic solution. Two three-phase transformers are introduced to improve this.
With a step-up transformer at the source, the voltage is increased, thereby reducing the current in the transmission lines since power loss...
With a step-up transformer at the source, the voltage is increased, thereby reducing the current in the transmission lines since power loss...
193
The Ideal Transformer
878
In single-phase two-winding transformers, two windings are coiled around a magnetic core characterized by cross-sectional area A and magnetic permeability μ. A phasor current i1 enters the left winding while i2 exits the right winding, establishing the fundamental working of the transformer through electromagnetic principles.
Ampere's Law forms the basis of understanding the magnetic field within the transformer. It states that the integral of the magnetic field intensity's...
Ampere's Law forms the basis of understanding the magnetic field within the transformer. It states that the integral of the magnetic field intensity's...
878

