Analog encoded neural network for power management in MPSoC

Larras, Benoît; Boguslawski, Bartosz; Lahuec, Cyril; Arzel, Matthieu; Seguin, Fabrice; Heitzmann, Frédéric

doi:10.1007/s10470-014-0420-z

Cited by 12 publications

(4 citation statements)

References 9 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…"Neural processors" circuits implement large Artificial Neural Networks (ANNs) composed of about one million neurons for embedded AI [7]- [9]. Oversized for tasks such as embedded associative memories, they are advantageously replaced by simpler clique-based ANNs [4], where specific classification tasks can be performed by networks of a few thousand of neurons in total [10]- [11]. Moreover, neuromorphic systems such as [12]- [13] operate at low frequency and are thus not adapted for real-time operation, unlike CBNNs which converge in tens of nanoseconds [14].…”

Section: Cbnn Structurementioning

confidence: 99%

On the Distribution of Clique-Based Neural Networks for Edge AI

Larras

Frappé

2020

IEEE J. Emerg. Sel. Topics Circuits Syst.

Self Cite

View full text Add to dashboard Cite

Distributed smart sensors are more and more used in applications such as biomedical or domestic monitoring. However, each sensor broadcasts data wirelessly to the others or to an aggregator, which leads to energy-hungry sensor nodes not ensuring data privacy. To tackle both challenges, this work proposes to distribute the feature extraction and a part of a clique-based neural network (CBNN) in each sensor node. This scheme allows standardizing data at the sensor level, ensuring privacy if the data is intercepted. Besides, a lower number of bits is transmitted, thus limiting the communication overhead. The inherent redundancy of clique-based networks makes them resilient to out-of-range connections, allowing an additional power reduction in the sensor nodes. Compared with a localized CBNN in the aggregator, the distributed structure reduces the inference latency by 28%, the sensor energy consumption by 25% and increases the protocol robustness. The circuit implementation is possible with the use of single-cluster iterative clique-based circuits, and demonstrated for a posture recognition application. To this end, a hardware circuit has been fabricated and performs a classification using 115fJ per synaptic event per neuron in 83ns.

show abstract

Section: Cbnn Structurementioning

confidence: 99%

On the Distribution of Clique-Based Neural Networks for Edge AI

Larras

Frappé

2020

IEEE J. Emerg. Sel. Topics Circuits Syst.

Self Cite

View full text Add to dashboard Cite

show abstract

“…As a result of the strong correlation brought by the connections of the clique embodying a message in the network, it is possible to retrieve the stored message based on partial or noisy information put into the network. In order to target first applications (Larras et al 2013a), the network was implemented in hardware (Larras et al 2013b).…”

Section: Message Retrieval Proceduresmentioning

confidence: 99%

Huffman Coding for Storing Non-Uniformly Distributed Messages in Networks of Neural Cliques

Boguslawski¹,

Gripon²,

Seguin³

et al. 2014

AAAI

Self Cite

View full text Add to dashboard Cite

Associative memories are data structures that allow retrieval of previously stored messages given part of their content. They thus behave similarly to human brain's memory that is capable for instance of retrieving the end of a song given its beginning. Among different families of associative memories, sparse ones are known to provide the best efficiency (ratio of the number of bits stored to that of bits used). Nevertheless, it is well known that non-uniformity of the stored messages can lead to dramatic decrease in performance. Recently, a new family of sparse associative memories achieving almost-optimal efficiency has been proposed. Their structure induces a direct mapping between input messages and stored patterns. In this work, we show the impact of non-uniformity on the performance of this recent model and we exploit the structure of the model to introduce several strategies to allow for efficient storage of non-uniform messages. We show that a technique based on Huffman coding is the most efficient.

show abstract

“…In [17] authors propose a power management technique for Multiprocessor System-on-Chip (MPSoC) based on the theory of [12]. Depending on the current working conditions and the data stored in the network the frequency for each Voltage/Frequency Island (VFI) is assigned.…”

Section: Application Examplementioning

confidence: 99%

“…This implies storing more cliques connected to a subgroup of fanals and produces local high density areas. By using the strategies proposed in the present work one can neutralize the influence of non-uniform data distribution and improve the performance of the technique proposed in [17].…”

Section: Application Examplementioning

confidence: 99%

Storing non-uniformly distributed messages in networks of neural cliques

Boguslawski¹,

Gripon²,

Seguin³

et al. 2013

Preprint

Self Cite

View full text Add to dashboard Cite

Associative memories are data structures that allow retrieval of stored messages from part of their content. They thus behave similarly to human brain that is capable for instance of retrieving the end of a song given its beginning.Among different families of associative memories, sparse ones are known to provide the best efficiency (ratio of the number of bits stored to that of bits used). Nevertheless, it is well known that non-uniformity of the stored messages can lead to dramatic decrease in performance. We introduce several strategies to allow efficient storage of non-uniform messages in recently introduced sparse associative memories. We analyse and discuss the methods introduced. We also present a practical application example.

show abstract

Analog encoded neural network for power management in MPSoC

Cited by 12 publications

References 9 publications

On the Distribution of Clique-Based Neural Networks for Edge AI

On the Distribution of Clique-Based Neural Networks for Edge AI

Huffman Coding for Storing Non-Uniformly Distributed Messages in Networks of Neural Cliques

Storing non-uniformly distributed messages in networks of neural cliques

Contact Info

Product

Resources

About