> ML_LITERATURE // IANDOLA-2016-SQUEEZENET-ALEXNET-ACCURACY-50X-FEWER-PARAMETERS_v1.0
SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <0.5MB model size
Forrest N. Iandola, Song Han, Matthew W. Moskewicz, Khalid Ashraf, William J. Dally, Kurt Keutzer · arXiv preprint (2016)
seminal-architecture2016foundationalthirdPartyReproduced
Principal Contribution
Designed Fire modules consisting of a squeeze 1x1 convolution layer and an expand 1x1/3x3 layer, achieving AlexNet accuracy with 50x fewer parameters and deep compression to 0.47 MB.
Operational Relevance
Serves as qualified reference for deploying task-image-classification in production.
Assumptions
- Spatial feature coherence and data manifold structure adhere to continuous representation hypotheses
Limitations
- Computational complexity scales with spatial resolution and parameter capacity
Connected Algorithms, Architectures & Tools
Related Algorithms:
Related Architectures:
Implementing Libraries: