14 October 2019

Photonic Computing paving the way to a "pervasive intelligence"

One of the most demanded tasks of AI is extracting patterns and features directly from collected big data. Among the various most promising approaches for accomplishing this goal, Deep Neural Networks (DNNs) are outperforming. 

The reason of the DNNs are so performing is not fully explained yet, but one possible explanation, widely elaborated in literature, is that being DNNs based on an iterative coarse-graining scheme, their functioning is somehow rooted to fundamental theoretical physics tool (e.g., Renormalization Group). 

The reverse side of the coin is that this is rather resource consuming and, as such, energy demanding. In fact, today DNNs (as other AI models) still rely on Boolean algebra transistors to do an enormous amount of computations over huge data sets. This has two major consequences: on one side chips and processors technologies aren’t getting faster at the same pace that AI methods and systems are progressing, and, on the other hand, current AI technologies are becoming more and more electricity-hungry. 

Today, for example, Cloud servers and data centers currently account for around 2% of power consumption in the U.S. According to some forecasts, data centers will consume one fifth of the world’s electricity by 2025. 

Will this energy consuming trend be really sustainable in long term scenarios (e.g., 6G) ?

Take a look at this paper - Lovén, Lauri, et al. "EdgeAI: A Vision for Distributed, Edge-native Artificial Intelligence in Future 6G Networks." The 1st 6G Wireless Summit (2019): 1-2.

We remind that in a DNN that each high-level layer learns increasingly abstract higher-level features, providing a useful, and at times reduced, representation of the features to a lower-level layer. This similarity, more specifically, is suggesting the intriguing possibility that DNNs principles are deeply rooted in quantum electromagnetics. This is offering a way to bypass above roadblocks: developing AI technologies based on photonic/optical computing systems which are faster and much less energy-hungry that current ones. 

Indeed, low-latency and low-energy neural network computations can be a game changer for a pervasive AI. In this direction, fully optical neural network could offer enhancements in computational speed and reduced power consumptions.

My last paper on these topics available at the following link:

09 October 2019

A pervasive "edge intelligence" ? Yes, but consuming less energy

After about five years of posts addressing various aspects on the evolution towards 5G Cloud-Edge Computing, I’ve been asked to start elaborating some ideas on “what’s next”. 

My take is that in the next 5-10 years there will be the true techno-economic chance of maturing and extending the perspective of the networks as part of a sort of pervasive “nervous system” of the Digital Society: a vision which I coined in 2014 for the first time at the Plenary of the EuCNC Conference in Bologna.

We know that a biological nervous system is a complex network of nerves and cells that carry messages to and from the brain and spinal cord to various parts of the body. The nervous system includes both the Central nervous system and Peripheral nervous system. The Central nervous system is made up of the brain and spinal cord and The Peripheral nervous system is made up of the Somatic and the Autonomic nervous systems.

Overall, we may summarize that a "nervous system” is about sensing the reality, comparing sensations with predictions and, eventually, acting on the reality in order to best adapt to the environment dynamics. This is a sort of “intelligence”, naturally embedded in living organisms. 

The idea that the brain, and more generally a nervous system, is like a network with inference engines is not new. As a matter of fact, the main task of the brain is trying to optimize probabilistic representations of what caused its sensory input: in other words, the brain has a model of the world that it tries to optimize it using sensory inputs to improve adaptation. This optimization is finessed using a (variational free-energy) bound on surprise.  

And this is done very efficiently, consuming only a few tenth of Watts !
This is great challenge as today AI is highly energy consuming ! Current AI technologies are very electricity-hungry, a problem that is manifesting itself both in the cloud and at the edge. Cloud servers and data centers currently account for around 2 percent of power consumption in the U.S. According to some forecasts, data centers will consume one fifth of the world’s electricity by 2025.

Take a look at this amazing paper by K. Friston, “The free-energy principle: a unified brain theory?” How can we bring these concepts into a pervasive network to transform it into a "nervous system"? 

In summary, it is likely we'll see a true “internet of intelligence” connecting “minds” with new forms of communications and interactions, sensing the reality with the most advance technologies (e.g., THz sensing), comparing sensations with predictions by means of Optical/Quantum Intelligence (much beyond today AI), and eventually, acting on the reality to best adapt to the environment dynamics.

31 July 2019

Complex Deep Learning with Quantum Optics

The rapid evolution towards future telecommunications infrastructures (e.g., 5G, the fifth generation of mobile networks) and the internet is renewing a strong interest for artificial intelligence (AI) methods, systems, and networks. 

Processing big data to infer patterns at high speeds and with low power consumption is becoming an increasing central technological challenge. Electronics are facing physically fundamental bottlenecks, whilst nanophotonics technologies are considered promising candidates to overcome the limitations of electronics. 

Today, there are evidences of an emerging research field, rooted in quantum optics, where the technological trajectories of deep neural networks (DNNs) and nanophotonics are crossing each other.

This paper elaborates on these topics and proposes a theoretical architecture for a Complex DNN made from programmable metasurfaces; an example is also provided showing a striking correspondence between the equivariance of convolutional neural networks (CNNs) and the invariance principle of gauge transformations.

Paper available at the following link:

12 March 2019

5G EdgeCloud - part 2

Cloudification” of telecommunications infrastructures and MEC/Edge-Cloud are leveraging on the exploitation of the (almost) full “virtualization” of both resources (e.g., processing, storage and networking) and network/service functions (e.g., Virtual Network Functions) up to the edge (i.e., access, distribution segments), or even beyond it up to the terminals or smart things (i.e., Fog Computing). 

MEC/Edge-Cloud is just a piece of the overall puzzle, as the border with Cloud Computing is disappearing soon !

This innovation trend in offering the opportunity of extending the business role of telecommunication Operators to play the role not only of Service Providers (for SaaS) but also the roles of both IaaS Infrastructure-as-a-Service (IaaS) Provider and Platform-as-a-Service (PaaS) Provider, in global markets.

This creates the conditions for boosting new open ecosystems, easing life to Applications and Service Developers.

In order to materialize this vision, there should be a clear definition of what are the services offered by the 5G IaaS and PaaS and the related open/standard APIs to access them.

For example, Infrastructure as a Service (IaaS) layer offers:
  • Raw virtual resources for connectivity, processing, storage;
  • Virtual Network Functions and other Network Services (e.g., middleboxes such as bridges, routers, load balancers, firewalls, video optimizers, etc).

Platform as a Service (PaaS) layer provides a controlled access - through standard interfaces API - to underneath IaaS. Moreover, PaaS include, for example:
  • Operating Systems services, services and apps development instruments and tools, database management capabilities, business analysis, A.I. tools, etc

The following picture is summarizing a number of possible business models, capable of re-defining the equilibria in the overall Industry.

08 March 2019

5G EdgeCloud - part 1

It was mid 2013 when we published one of the first visionary papers on EdgeCloud, in the IEEE Communications Magazine. 

After six years, MEC/EdgeCloud indefinitely under the spot in industry, worldwide.  In fact, MEC/EdgeCloud is expected to play a key strategic role in this Digital Transformation towards 5G.

There are evidences that, today, Network and Service Providers are exploring different strategies for MEC/Edge Cloud introduction and exploitation, mainly (but not only) motivated by the potential opportunities for: (i) saving costs in the Digital Transformation of the network and service infrastructures; (ii) generating new revenues, e.g., by improving performance of current services and enabling new ones, with the related business models.

These topics are addressed by several standardization bodies and fora.
A non- exhaustive list includes:
Telecom Infra Project (WGs on Edge Computing)
EdgeX Foundry
Open Edge Computing
In general, the overall standardization picture is rather fragmented but these bodies are addressing MEC/Edge from different perspectives (they are not fully overlapping) and there is a common awareness that global interoperability is a “must” for enabling new services ecosystems.

This means that Industry need to align on open and common APIs capable to ease Service and Apps Developers: in fact, this is crucial to promotes innovation and accelerates development by Third Parties applications and services, capable of enabling Network and Service Providers to capitalize on their investments on EdgeCloud.