Adaptive Neural Networks for Efficient Inference

Authors: Tolga Bolukbasi, Joseph Wang, Ofer Dekel, Venkatesh Saligrama

ICML 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Empirically, these approaches yield dramatic reductions in computational cost, with up to a 2.8x speedup on state-of-the-art networks from the Image Net image recognition challenge with minimal (< 1%) loss of top5 accuracy.
Researcher Affiliation Collaboration 1Boston University, Boston, MA, USA 2Amazon, Cambridge, MA, USA 3Microsoft Research, Redmond, WA, USA.
Pseudocode Yes Algorithm 1 Adaptive Network Learning Pseudocode
Open Source Code No The paper does not provide any statement or link regarding the availability of its source code.
Open Datasets Yes We evaluate our method on the Imagenet 2012 classification dataset (Russakovsky et al., 2015) which has 1000 object classes. We train using the 1.28 million training images and evaluate the system using 50k validation images.
Dataset Splits Yes We train using the 1.28 million training images and evaluate the system using 50k validation images.
Hardware Specification Yes We measure network times using the built-in tool in the Caffe library on a server that utilizes a Nvidia Titan X Pascal with Cu DNN 5.
Software Dependencies Yes We measure network times using the built-in tool in the Caffe library on a server that utilizes a Nvidia Titan X Pascal with Cu DNN 5.
Experiment Setup Yes To output a prediction following each convolutional layer, we train a single layer linear classifier after a global average pooling for each layer... We sweep the cost trade-off parameter in the range 0.0 to 0.1 to achieve different budget points.