TRAINING DEEP NEURAL NETWORKS WITH LOW PRECISION MULTIPLICATIONS
文章：https://arxiv.org/abs/1412.7024 0 摘要 乘法器是深度神經網絡數字實現中空間和功耗最大的算術運算符。我們在三個基準數據集上訓練了一套最先進的神經網絡（Maxout網絡）：MNIST，CIFAR-10和SVHN。它們採用三種不同的格式進行訓練：浮點，固定點和動態固定點。對於每個數據集以及每種格式，我們評估乘法的精確度對訓練後最終誤差的影響。我們發現，非常低的精度
相關文章
- 1. Training With Mixed Precision
- 2. 《SWALP：Stochastic Weight Averaging in Low-Precision Training》
- 3. Learning Accurate Low-Bit Deep Neural Networks with Stochastic Quantization
- 4. BinaryConnect： Training Deep Neural Networks with binary weights during propagations
- 5. DOREFA-NET: TRAINING LOW BITWIDTH CONVOLUTIONAL NEURAL NETWORKS WITH LOW BITWIDTH GRADIENTS
- 6. CHAPTER 11-Training Deep Neural Nets-part3
- 7. INQ(incremental network quantization：towards lossless CNNs with low-precision weights
- 8. 論文翻譯：Deep Learning with Low Precision by Half-wave Gaussian Quantization
- 9. Tips for Training Deep Neural Network
- 10. Accelerating deep convolutional networks using low-precision and sparsity
- 更多相關文章...
- • XSLT
- • Docker 容器連接 - Docker教程
- • 爲了進字節跳動，我精選了29道Java經典算法題，帶詳細講解
- • 算法總結-股票買賣
相關標籤/搜索
每日一句
-
每一个你不满意的现在，都有一个你没有努力的曾经。
最新文章
歡迎關注本站公眾號,獲取更多信息
相關文章
- 1. Training With Mixed Precision
- 2. 《SWALP：Stochastic Weight Averaging in Low-Precision Training》
- 3. Learning Accurate Low-Bit Deep Neural Networks with Stochastic Quantization
- 4. BinaryConnect： Training Deep Neural Networks with binary weights during propagations
- 5. DOREFA-NET: TRAINING LOW BITWIDTH CONVOLUTIONAL NEURAL NETWORKS WITH LOW BITWIDTH GRADIENTS
- 6. CHAPTER 11-Training Deep Neural Nets-part3
- 7. INQ(incremental network quantization：towards lossless CNNs with low-precision weights
- 8. 論文翻譯：Deep Learning with Low Precision by Half-wave Gaussian Quantization
- 9. Tips for Training Deep Neural Network
- 10. Accelerating deep convolutional networks using low-precision and sparsity