X1000 real-time phoneme recognition VLSI using feed-forward deep neural networks

By Jonghong Kim, Kyuyeon Hwang, Wonyong Sung. 2014

55 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Wonyong Sung
2. Daniel Soudry
3. Sungho Shin
4. Itay Hubara
5. Yoshua Bengio
6. Sajid Anwar
7. Song Han
8. Martin Jaggi

In Computer Vision and Pattern Recognition

1. Wonyong Sung
2. Daniel Soudry
3. Yoshua Bengio
4. Sajid Anwar
5. Song Han
6. Itay Hubara
7. Luc Van Gool
8. Sungho Shin

Citations in the corpus

• Binarized Neural Networks (2016, methodology): “tions avoided by the proposed BNN algorithm. Hwang & Sung (2014) focused on a fixed-point neural network design and achieved performance almost identical to that of the floating-point architecture. Kim et al. (2014) provided evidence that DNNs with ternary weights, used on a dedicated circuit, con”
• Fixed-point feedforward deep neural network design using weights +1, 0, and −1 (2014, methodology): “In this paper, we propose a high performance fixed-point optimization method that can greatly reduce the word-length of weights and signals for implementing DNNs.”
• BinaryConnect: Training Deep Neural Networks with binary weights during propagations (2015, background): “Training DNNs with binary weights has been the subject of very recent works [37, 38, 39, 40].”
• A comprehensive survey on model compression and acceleration (2020, methodology): “Kim et al. (2014) presented VLSI based implementation of DNN for phoneme recognition in which they represented each weight with +Δ , 0 or −Δ .”
• Deep Learning with Limited Numerical Precision (2015, background): “, 2014), or take the route of custom hardware implementations (Kim et al., 2014; Merolla et al., 2014).”
• Stripes: Bit-serial deep neural network computing (2016, methodology): “[29] use reduced precision neurons (8 and 4 bits) and weights (2 bits) to reduce the circuit size and memory requirements of a fixed hardware”
• Fixed point optimization of deep convolutional neural networks for object recognition (2015, background): “A quantized feed forward deep neural network is introduced in [8] and [9].”
• BinaryNet: Training Deep Neural Networks with Weights and Activations Constrained to +1 or -1 (2016, background): “Hwang & Sung (2014); Kim et al. (2014) retrain neural networks with ternary weights and 3-bit activations, i.e.: • They train a neural network with high-precision,”
• Resiliency of Deep Neural Networks under Quantization (2015, methodology): “The ternary weight based FFDNN and CNN are used for VLSI and FPGA based implementations, by which the algorithms can operate with only on-chip memory consuming very low power (Kim et al., 2014).”