The edge processing of deep neural networks (DNNs) is becoming increasingly
important due to its ability to extract valuable information directly at the
data source to minimize latency and energy consumption. Frequency-domain model
compression, such as with the Walsh-Hadamard transform (WHT), has been
identified as an efficient alternative. However, the benefits of
frequency-domain processing are often offset by the increased
multiply-accumulate (MAC) operations required. This paper proposes a novel
approach to an energy-efficient acceleration of frequency-domain neural
networks by utilizing analog-domain frequency-based tensor transformations. Our
approach offers unique opportunities to enhance computational efficiency,
resulting in several high-level advantages, including array micro-architecture
with parallelism, ADC/DAC-free analog computations, and increased output
sparsity. Our approach achieves more compact cells by eliminating the need for
trainable parameters in the transformation matrix. Moreover, our novel array
micro-architecture enables adaptive stitching of cells column-wise and
row-wise, thereby facilitating perfect parallelism in computations.
Additionally, our scheme enables ADC/DAC-free computations by training against
highly quantized matrix-vector products, leveraging the parameter-free nature
of matrix multiplications. Another crucial aspect of our design is its ability
to handle signed-bit processing for frequency-based transformations. This leads
to increased output sparsity and reduced digitization workload. On a
16$\times$16 crossbars, for 8-bit input processing, the proposed approach
achieves the energy efficiency of 1602 tera operations per second per Watt
(TOPS/W) without early termination strategy and 5311 TOPS/W with early
termination strategy at VDD = 0.8 V.

通过利用模拟域基于频率的张量变换，本文提出了一种能够实现能耗高效的频率域神经网络加速方法，包括边缘处理、频域模型压缩、数组微架构和能效提升等关键技术。

无需 ADC/DAC 的深度神经网络频率转换模拟加速

ADC/DAC-Free Analog Acceleration of Deep Neural Networks with Frequency  Transformation

Internet of Things (IoT) devices and applications are being deployed in our
homes and workplaces. These devices often rely on continuous data collection to
feed machine learning models. However, this approach introduces several privacy
and efficiency challenges, as the service operator can perform unwanted
inferences on the available data. Recently, advances in edge processing have
paved the way for more efficient, and private, data processing at the source
for simple tasks and lighter models, though they remain a challenge for larger,
and more complicated models. In this paper, we present a hybrid approach for
breaking down large, complex deep neural networks for cooperative,
privacy-preserving analytics. To this end, instead of performing the whole
operation on the cloud, we let an IoT device to run the initial layers of the
neural network, and then send the output to the cloud to feed the remaining
layers and produce the final result. In order to ensure that the user's device
contains no extra information except what is necessary for the main task and
preventing any secondary inference on the data, we introduce Siamese
fine-tuning. We evaluate the privacy benefits of this approach based on the
information exposed to the cloud service. We also assess the local inference
cost of different layers on a modern handset. Our evaluations show that by
using Siamese fine-tuning and at a small processing cost, we can greatly reduce
the level of unnecessary, potentially sensitive information in the personal
data, and thus achieving the desired trade-off between utility, privacy, and
performance.

本篇论文提出了一种用于合作、隐私保护分析的深度神经网络混合方法，使用边缘处理的思想，在保持效益的同时，有效减少了不必要、潜在敏感信息的级别。利用 Siamese 微调技术，可以确保用户设备只带有主任务所需的必要信息，并防止数据的二次推断。

一个混合深度学习架构用于隐私保护移动分析

A Hybrid Deep Learning Architecture for Privacy-Preserving Mobile  Analytics

This work studies the joint design of cloud and edge processing for the
downlink of a fog radio access network (F-RAN). In an F-RAN, as in cloud-RAN
(C-RAN), a baseband processing unit (BBU) can perform joint baseband processing
on behalf of the remote radio heads (RRHs) that are connected to the BBU by
means of the fronthaul links. In addition to the minimal functionalities of
conventional RRHs in C-RAN, the RRHs in an F-RAN may be equipped with local
caches, in which frequently requested contents can be stored, as well as with
baseband processing capabilities. They are hence referred to as enhanced RRH
(eRRH). This work focuses on the design of the delivery phase for an arbitrary
pre-fetching strategy used to populate the caches of the eRRHs. Two
fronthauling modes are considered, namely a hard-transfer mode, whereby
non-cached files are communicated over the fronthaul links to a subset of
eRRHs, and a soft-transfer mode, whereby the fronthaul links are used to convey
quantized baseband signals as in a C-RAN. Unlike the hard-transfer mode in
which baseband processing is traditionally carried out only at the eRRHs, the
soft-transfer mode enables both centralized precoding at the BBU and local
precoding at the eRRHs based on the cached contents, by means of a novel
superposition coding approach. To attain the advantages of both approaches, a
hybrid design of soft- and hard-transfer modes is also proposed. The problem of
maximizing the delivery rate is tackled under fronthaul capacity and per-eRRH
power constraints. Numerical results are provided to compare the performance of
hard- and soft-transfer fronthauling modes, as well as of the hybrid scheme,
for different baseline pre-fetching strategies.

本文研究了云和边缘处理的联合设计，主要集中在设计优化 FRAN（Fog Radio Access Network）中的缓存管理和前传设计等核心问题，以最大化传输速率。