Abstract:Machine-learning algorithms have shown outstanding image recognition/classification performance for computer vision applications. However, the compute and energy requirement for implementing such classifier models for large-scale problems is quite high. In this paper, we propose feature driven selective classification (FALCON) inspired by the biological visual attention mechanism in the brain to optimize the energy-efficiency of machine-learning classifiers. We use the consensus in the characteristic features (color/texture) across images in a dataset to decompose the original classification problem and construct a tree of classifiers (nodes) with a generic-to-specific transition in the classification hierarchy. The initial nodes of the tree separate the instances based on feature information and selectively enable the latter nodes to perform object specific classification. The proposed methodology allows selective activation of only those branches and nodes of the classification tree that are relevant to the input while keeping the remaining nodes idle. Additionally, we propose a programmable and scalable neuromorphic engine (NeuE) that utilizes arrays of specialized neural computational elements to execute the FALCON-based classifier models for diverse datasets. The structure of FALCON facilitates the reuse of nodes while scaling up from small classification problems to larger ones thus allowing us to construct classifier implementations that are significantly more efficient. We evaluate our approach for a 12-object classification task on the Caltech101 dataset and ten-object task on CIFAR-10 dataset by constructing FALCON models on the NeuE platform in 45-nm technology. Our results demonstrate up to $3.66oldsymbol imes $ improvement in energy-efficiency for no loss in output quality, and even higher improvements of up to $5.91oldsymbol imes $ with 3.9% accuracy loss compared to an optimized baseline network. In addition, FALCON shows an improvement in training time of up to $1.96oldsymbol imes $ as compared to the traditional classification approach.

BFALCON Generalization Capability Improvement Based on PCA Initialization

Sigmoid membership function based FALCON and its application

Neural Incremental Attribute Learning Based on Principal Component Analysis

Fuzzy adaptive learning control network with sigmoid membership funcflon

On the Improvement of Generalization and Stability of Forward-Only Learning via Neural Polarization

FALCON: Feedback-driven Adaptive Long/short-term memory reinforced Coding Optimization system

A fuzzy logic controller designed based on CPN and its application

Robust Principal Component Analysis Based on Fuzzy Local Information Reservation

Target Recognition in Naval Battlefield Based on Principal Component Analysis and Neural Networks

An adaptive learning algorithm for principal component analysis

Fine-grained Classes and How to Find Them

Identifying Feedforward and Feedback Controllable Subspaces of Neural Population Dynamics

A genetic algorithm for rule extraction in fuzzy adaptive learning control networks

FL-AGCNS: Federated Learning Framework for Automatic Graph Convolutional Network Search

PCANet: A Simple Deep Learning Baseline for Image Classification?

Fuzzy advantage learning

FALCON: Feature Driven Selective Classification for Energy-Efficient Image Recognition

A New Method for Decision on the Structure of RBF Neural Network

FALCON: FLOP-Aware Combinatorial Optimization for Neural Network Pruning

PCA-Boosted Autoencoders for Nonlinear Dimensionality Reduction in Low Data Regimes

Factorization of Broad Expansion for Broad Learning System