Abstract
The number of groups ($g$) in group convolution (GConv) is selected to boost the predictive performance of deep neural networks (DNNs) in a compute and parameter efficient manner. However, we show that naive selection of $g$ in GConv creates an imbalance between the computational complexity and degree of data reuse, which leads to suboptimal energy efficiency in DNNs. We devise an optimum group size model, which enables a balance between computational cost and data movement cost, thus, optimize the energy-efficiency of DNNs. Based on the insights from this model, we propose an "energy-efficient group convolution" (E2GC) module where, unlike the previous implementations of GConv, the group size ($G$) remains constant. Further, to demonstrate the efficacy of the E2GC module, we incorporate this module in the design of MobileNet-V1 and ResNeXt-50 and perform experiments on two GPUs, P100 and P4000. We show that, at comparable computational complexity, DNNs with constant group size (E2GC) are more energy-efficient than DNNs with a fixed number of groups (FgGC). For example, on P100 GPU, the energy-efficiency of MobileNet-V1 and ResNeXt-50 is increased by 10.8% and 4.73% (respectively) when E2GC modules substitute the FgGC modules in both the DNNs. Furthermore, through our extensive experimentation with ImageNet-1K and Food-101 image classification datasets, we show that the E2GC module enables a trade-off between generalization ability and representational power of DNN. Thus, the predictive performance of DNNs can be optimized by selecting an appropriate $G$.
Introduction
As deep learning moves toward battery-constrained edge devices such as drones, wearables, and self-driving cars, optimizing for "FLOPs" or "Parameters" is no longer sufficient. We must optimize for energy. Standard Group Convolutions (GConv) and Depthwise Convolutions (DWConv) are popular because they drastically reduce Multiply-Accumulate (MAC) operations. However, there is a hidden cost: as the number of groups ($g$) increases to save on computation, data reuse simultaneously decreases. When data reuse drops, the system spends more energy moving data from memory to the processor than it does performing arithmetic operations. Most existing architectures fix the number of groups ($g$) across the network, leading to suboptimal energy usage and fragmented memory access.
Balancing Computation and Data Movement
The authors propose a shift in perspective: instead of fixing the number of groups ($g$), we should fix the group size ($G$)—the number of channels within each group. This approach, termed Energy-efficient Group Convolution (E2GC), balances the trade-off between reduced computation and the high cost of data movement.
The researchers derived a mathematical model to find the "sweet spot" for energy efficiency, accounting for the disparity between arithmetic operation costs and memory access costs. In the E2GC module, as the network gets deeper and the number of channels ($m$) increases, the number of groups ($g$) also increases to keep $G$ (where $G = m/g$) constant. This maintains a consistent Arithmetic Intensity (AI) across layers, keeping hardware utilization high.
Experimental Results
The E2GC module was evaluated by replacing standard layers in MobileNet-V1 and ResNeXt-50 on NVIDIA P100 and P4000 GPUs. The comparison was made against "Fixed group Group Convolution" (FgGC), the standard industry practice of keeping $g$ constant.
On the P100 GPU, E2GC increased the energy efficiency of MobileNet-V1 by 10.8% and ResNeXt-50 by 4.73% compared to standard fixed-group implementations. On the P4000 GPU, MobileNet-V1 with E2GC ($G=32$) was up to 65.4% more energy-efficient than the baseline. Beyond energy savings, E2GC often improved accuracy. For ResNeXt-50 on ImageNet, the E2GC version reached 77.04% Top-1 accuracy, significantly higher than the baseline's 74.43% at comparable computational complexity.
Conclusion
E2GC demonstrates that fixing group size rather than the number of groups leads to more energy-efficient DNN designs. While the optimal group size $G$ is platform-dependent, the E2GC framework provides a robust method for balancing generalization ability and representational power. Future work may explore the application of E2GC to other tasks like object detection and segmentation.
Summary updated