Agenda
MSc ME Thesis Presentation
- Tuesday, 29 November 2022
- 15:00-15:40
- HB17.140
A New Logarithmic Quantization Technique and Corresponding Processing Element Design for CNN Accelerators
Longxing JiangConvolutional Neural Networks (CNN) have become a popular solution for computer vision problems. However, due to the high data volumes and intensive computation involved in CNNs, deploying CNNs on low-power hardware systems is still challenging. The power consumption of CNNs can be prohibitive in the most common implementation platforms: CPUs and GPUs. Therefore, hardware accelerators that can exploit CNN parallelism and methods to reduce the computation burden or memory requirements are still hot research topics. Quantization is one of these methods.
One suitable quantization strategy for low-power deployments is logarithmic quantization.
Logarithmic quantization for Convolutional Neural Networks (CNN): a) fits well typical weights and activation distributions, and b) allows the replacement of the multiplication operation by a shift operation that can be implemented with fewer hardware resources. In this thesis, a new quantization method named Jumping Log Quantization (JLQ) is proposed. The key idea of JLQ is to extend the quantization range, by adding a coefficient parameter ”s” in the power of two exponents (2sx+i ).
This quantization strategy skips some values from the standard logarithmic quantization. In addition, a small hardware-friendly optimization called weight de-zeroing is proposed in this work. Zero-valued weights that cannot be performed by a single shift operation are all replaced with logarithmic weights to reduce hardware resources with little accuracy loss.
To implement the Multiply-And-Accumulate (MAC) operation (needed to compute convolutions) when the weights are JLQ-ed and dezeroed, a new Processing Element (PE) have been developed. This new PE uses a modified barrel shifter that can efficiently avoid the skipped values.
Resource utilization, area, and power consumption of the new PE standing alone and in a systolic array prototype are reported. The results show that JLQ performs better than other state-of-the-art logarithmic quantization methods when the bit width of the operands becomes very small.
Agenda
- Thu, 25 Apr 2024
- 11:00
- HB 17.140
Signal Processing Seminar
Yanbin He
Modelling Error Correction in Sparse Bayesian Learning via Grid Optimization
- Tue, 30 Apr 2024
- 10:00
- HB18.090
MSc SPS Thesis presentation
Wim Kok
A SystemC SNN model for power trace generation
- Mon, 6 May 2024
- 12:30
- Aula Senaatszaal
PhD Thesis Defence
Christoph Manss
Multi-agent exploration under sparsity constraints
- Tue, 21 May 2024
- 10:00
- Aula Senaatszaal
PhD Thesis Defence
Wangyang Yu
- 27 -- 28 May 2024
- Aula, TU Delft
Conferences
44th Benelux Symposium on Information Theory and Signal Processing (SITB'24, Delft)
- Tue, 18 Jun 2024
- 15:00
- Aula Senaatszaal
PhD Thesis Defence
Hanie Moghaddasi
Model-based feature engineering of atrial fibrillation
- Mon, 24 Jun 2024
- Aula, TU Delft