Connect with us

Security

Google’s DeepMind asks what it means for AI to fail

Published

on

There’s been years of study placed in the problem of how to make artificial intelligence “robust” to attack and less prone to failure. Yet the field is still coming to grips with what failure in AI actually means, as pointed out by a blog post this week from the DeepMind unit of Google.

The missing element may seem obvious to some: it would really help if there was more human involvement in setting the boundary conditions for how neural networks are supposed to function.

Researchers Pushmeet Kohli, Sven Gowal, Krishnamurthy, Dvijotham, and Jonathan Uesato have been studying the problem, and they identify much work that remains to be done, which they sum up under the title “Towards Robust and Verified AI: Specification Testing, Robust Training, and Formal Verification.”

There’s a rich history of verification testing for computer programs, but those approaches are “not not suited for modern deep learning systems.” 

Also: MIT ups the ante in getting one AI to teach another 

Why? In large part because scientists are still learning about what it means for a neural network to follow the “specification” that was laid out for it. It’s not always clear what the specification even is.

“Specifications that capture ‘correct’ behavior in AI systems are often difficult to precisely state,” the authors write. 

Google’s DeepMind proposes ways to set a bound on the kinds of outputs a neural network can produce, to keep it from doing the wrong thing. 


DeepMind

The notion of a “specification” comes out of the software world, the DeepMind researchers observe. It is the intended functionality of a computer system. 

As the authors wrote in a post in December, in AI, there may not be just one spec, there may be at least three. There is the “ideal” specification, what the system’s creators imagine it could do. Then there is the “design” specification, the “objective function” explicitly optimized for a neural network. And, lastly, there is the “revealed” specification, the way that the thing actually performs. They call these three specs, which all can vary quite a bit from one another, the wish, the design, and the behavior. 

Designing artificial neural networks can be seen as how to close the gap between wish, design and behavior. As they wrote in the December essay, “A specification problem arises when there is a mismatch between the ideal specification and the revealed specification, that is, when the AI system doesn’t do what we’d like it to do. ”

Also: Google ponders the shortcomings of machine learning

They propose various routes to test and train neural networks that are more robust to errors, and presumably more faithful to specs. 

One approach is to use AI itself to figure out what befuddles AI. That means using a reinforcement learning system, like Google’s AlphaGo, to find the worst possible ways that another reinforcement learning system can fail? 

The authors did just that, in a paper published in December. “We learn an adversarial value function which predicts from experience which situations are most likely to cause failures for the agent.” The agent in this case refers to a reinforcement learning agent. 

“We then use this learned function for optimisation to focus the evaluation on the most problematic inputs.” They claim that the method leads to “large improvements over random testing” of reinforcement learning systems.

Must read


Another approach is to train a neural network to avoid a whole range of outputs, to keep it from going entirely off the rails and making really bad predictions. The authors claim that a “simple bounding technique,” something called “interval bound propagation,” is capable of training a “verifiably robust” neural network. That work won them a “best paper” award at the NeurIPS conference last year.

They’re now moving beyond just testing and training a neural network to avoid disaster, they’re also starting to find a theoretical basis for a guarantee of robustness. They approached it as an “optimisation problem that tries to find the largest violation of the property being verified.” 

Despite those achievements, at the end of the day, “much work is needed,” the authors write “to build automated tools for ensuring that AI systems in the real world will do the ‘right thing’.” 

Some of that work is to design algorithms that can test and train neural networks more intensely. But some of it probably involves a human element. It’s about setting the goals — the objective function — for AI that matches what humans want. 

“Building systems that can use partial human specifications and learn further specifications from evaluative feedback would be required,” they write, “as we build increasingly intelligent agents capable of exhibiting complex behaviors and  acting in unstructured environments.”

Previous and related coverage:

What is AI? Everything you need to know

An executive guide to artificial intelligence, from machine learning and general AI to neural networks.

What is deep learning? Everything you need to know

The lowdown on deep learning: from how it relates to the wider field of machine learning through to how to get started with it.

What is machine learning? Everything you need to know

This guide explains what machine learning is, how it is related to artificial intelligence, how it works and why it matters.

What is cloud computing? Everything you need to know about

An introduction to cloud computing right from the basics up to IaaS and PaaS, hybrid, public, and private cloud.

Related stories:

Source link

Continue Reading
Click to comment

Leave a Reply

Your email address will not be published.

Security

Retrospective thoughts on KubeCon Europe 2022

Published

on

I’m not going to lie. As I sit on a plane flying away from Valencia, I confess to have been taken aback by the scale of Kubecon Europe this year. In my defence, I wasn’t alone the volume of attendees appeared to take conference organisers and exhibitors by surprise, illustrated by the notable lack of water, (I was told) t-shirts and (at various points) taxis.

Keynotes were filled to capacity, and there was a genuine buzz from participants which seemed to fall into two camps: the young and cool, and the more mature and soberly dressed.

My time was largely spent in one-on-one meetings, analyst/press conferences and walking the stands, so I can’t comment on the engineering sessions. Across the piece however, there was a genuine sense of Kubernetes now being about the how, rather than the whether. For one reason or another, companies have decided they want to gain the benefits of building and deploying distributed, container-based applications.

Strangely enough, this wasn’t being seen as some magical sword that can slay the dragons of legacy systems and open the way to digital transformation the kool-aid was as absent as the water. Ultimately, enterprises have accepted that, from an architectural standpoint and for applications in general, the Kubernetes model is as good as any available right now, as a non-proprietary, well-supported open standard that they can get behind.

Virtualisation-based options and platform stacks are too heavyweight; serverless architectures are more applicable to specific use cases. So, if you want to build an application and you want it to be future-safe, the Kubernetes target is the one to aim for.

Whether to adopt Kubernetes might be a done deal, but how to adopt certainly is not. The challenge is not with Kubernetes itself, but everything that needs to go around it to make resulting applications enterprise-ready.

For example, they need to operate in compliance environments; data needs to be managed, protected, and served into an environment that doesn’t care too much about the state; integration tools are required with external and legacy systems; development pipelines need to be in place, robust and value-focused; IT Operations need a clear view of what’s running whereas a bill of materials, and the health of individual clusters; and disaster recovery is a must.

Kubernetes doesn’t do these things, opening the door to an ecosystem of solution vendors and (often CNCF-backed) open source projects. I could drill into these areas Service Mesh, GitOps, orchestration, observability, and backup but the broader point is that they are all evolving and coalescing around the need. As they increase in capability, barriers to adoption reduce and the number of potential use cases grows.

All of which puts the industry at an interesting juncture. It’s not that tooling isn’t ready: organizations are already successfully deploying applications based on Kubernetes. In many cases, however, they are doing more work than they need developers need insider knowledge of target environments, interfaces need to be integrated rather than using third-party APIs, higher-order management tooling (such as AIOps) has to be custom-deployed rather than recognising the norms of Kubernetes operations.

Solutions do exist, but they tend to be coming from relatively new vendors that are feature rather than platform players, meaning that end-user organisations have to choose their partners wisely, then build and maintain development and management platforms themselves rather than using pre-integrated tools from a singe vendor.

None of this is a problem per se, but it does create overheads for adopters, even if they gain earlier benefits from adopting the Kubernetes model. The value of first-mover advantage has to be weighed against that of investing time and effort in the current state of tooling: as a travel company once told me, “we want to be the world’s best travel site, not the world’s best platform engineers.”

So, Kubernetes may be inevitable, but equally, it will become simpler, enabling organisations to apply the architecture to an increasingly broad set of scenarios. For organisations yet to make the step towards Kubernetes, now may still be a good time to run a proof of concept though in some ways, that sip has sailed perhaps focus the PoC on what it means for working practices and structures, rather than determining whether the concepts work at all.

Meanwhile and perhaps most importantly, now is a very good moment for organisations to look for what scenarios Kubernetes works best “out of the box”, working with providers and reviewing architectural patterns to deliver proven results against specific, high-value needs these are likely to be by industry and by the domain (I could dig into this, but did I mention that I’m sitting on a plane? ).

Jon Collins from Kubecon 2022

Kubernetes might be a done deal, but that doesn’t mean it should be adopted wholesale before some of the peripheral detail is ironed out.

The post Retrospective thoughts on KubeCon Europe 2022 appeared first on GigaOm.

Continue Reading

Security

Retrospective thoughts on Kubecon

Published

on

I’m not going to lie. As I sit on a plane flying away from Valencia, I confess to have been taken aback by the scale of Kubecon Europe this year. In my defence, I wasn’t alone the volume of attendees appeared to take conference organisers and exhibitors by surprise, illustrated by the notable lack of water, (I was told) t-shirts and (at various points) taxis.

Keynotes were filled to capacity, and there was a genuine buzz from participants which seemed to fall into two camps: the young and cool, and the more mature and soberly dressed.

My time was largely spent in one-on-one meetings, analyst/press conferences and walking the stands, so I can’t comment on the engineering sessions. Across the piece however, there was a genuine sense of Kubernetes now being about the how, rather than the whether. For one reason or another, companies have decided they want to gain the benefits of building and deploying distributed, container-based applications.

Strangely enough, this wasn’t being seen as some magical sword that can slay the dragons of legacy systems and open the way to digital transformation the kool-aid was as absent as the water. Ultimately, enterprises have accepted that, from an architectural standpoint and for applications in general, the Kubernetes model is as good as any available right now, as a non-proprietary, well-supported open standard that they can get behind.

Virtualisation-based options and platform stacks are too heavyweight; serverless architectures are more applicable to specific use cases. So, if you want to build an application and you want it to be future-safe, the Kubernetes target is the one to aim for.

Whether to adopt Kubernetes might be a done deal, but how to adopt certainly is not. The challenge is not with Kubernetes itself, but everything that needs to go around it to make resulting applications enterprise-ready.

For example, they need to operate in compliance environments; data needs to be managed, protected, and served into an environment that doesn’t care too much about the state; integration tools are required with external and legacy systems; development pipelines need to be in place, robust and value-focused; IT Operations need a clear view of what’s running whereas a bill of materials, and the health of individual clusters; and disaster recovery is a must.

Kubernetes doesn’t do these things, opening the door to an ecosystem of solution vendors and (often CNCF-backed) open source projects. I could drill into these areas Service Mesh, GitOps, orchestration, observability, and backup but the broader point is that they are all evolving and coalescing around the need. As they increase in capability, barriers to adoption reduce and the number of potential use cases grows.

All of which puts the industry at an interesting juncture. It’s not that tooling isn’t ready: organizations are already successfully deploying applications based on Kubernetes. In many cases, however, they are doing more work than they need developers need insider knowledge of target environments, interfaces need to be integrated rather than using third-party APIs, higher-order management tooling (such as AIOps) has to be custom-deployed rather than recognising the norms of Kubernetes operations.

Solutions do exist, but they tend to be coming from relatively new vendors that are feature rather than platform players, meaning that end-user organisations have to choose their partners wisely, then build and maintain development and management platforms themselves rather than using pre-integrated tools from a singe vendor.

None of this is a problem per se, but it does create overheads for adopters, even if they gain earlier benefits from adopting the Kubernetes model. The value of first-mover advantage has to be weighed against that of investing time and effort in the current state of tooling: as a travel company once told me, “we want to be the world’s best travel site, not the world’s best platform engineers.”

So, Kubernetes may be inevitable, but equally, it will become simpler, enabling organisations to apply the architecture to an increasingly broad set of scenarios. For organisations yet to make the step towards Kubernetes, now may still be a good time to run a proof of concept though in some ways, that sip has sailed perhaps focus the PoC on what it means for working practices and structures, rather than determining whether the concepts work at all.

Meanwhile and perhaps most importantly, now is a very good moment for organisations to look for what scenarios Kubernetes works best “out of the box”, working with providers and reviewing architectural patterns to deliver proven results against specific, high-value needs these are likely to be by industry and by the domain (I could dig into this, but did I mention that I’m sitting on a plane? ).

Jon Collins from Kubecon 2022

Kubernetes might be a done deal, but that doesn’t mean it should be adopted wholesale before some of the peripheral detail is ironed out.

The post Retrospective thoughts on Kubecon appeared first on GigaOm.

Continue Reading

Security

Defeating Distributed Denial of Service Attacks

Published

on

It seems like every day the news brings new stories of cyberattacks. Whether ransomware, malware, crippling viruses, or more frequently of late—distributed denial of service (DDoS) attacks. According to Infosec magazine, in the first half of 2020, there was a 151% increase in the number of DDoS attacks compared to the same period the previous year. That same report states experts predict as many as 15.4 million DDoS attacks within the next two years.

These attacks can be difficult to detect until it’s too late, and then they can be challenging to defend against. There are solutions available, but there is no one magic bullet. As Alastair Cooke points out in his recent “GigaOm Radar for DDoS Protection” report, there are different categories of DDoS attacks.

And different types of attacks require different types of defenses. You’ll want to adopt each of these three defense strategies against DDoS attacks to a certain degree, as attackers are never going to limit themselves to a single attack vector:

Network Defense: Attacks targeting the OS and network operate at either Layer 3 or Layer 4 of the OSI stack. These attacks don’t flood the servers with application requests but attempt to exhaust TCP/IP resources on the supporting infrastructure. DDoS protection solutions defending against network attacks identify the attack behavior and absorb it into the platform.

Application Defense: Other DDoS attacks target the actual website itself or the web server application by overwhelming the site with random data and wasting resources. DDoS protection against these attacks might handle SSL decryption with hardware-based cryptography and prevent invalid data from reaching web servers.

Defense by Scale: There have been massive DDoS attacks, and they show no signs of stopping. The key to successfully defending against a DDoS attack is to have a scalable platform capable of deflecting an attack led by a million bots with hundreds of gigabits per second of network throughput.

Table 1. Impact of Features on Metrics
[chart id=”1001387″ show=”table”]

DDoS attacks are growing more frequent and more powerful and sophisticated. Amazon reports mitigating a massive DDoS attack a couple of years ago in which peak traffic volume reached 2.3 Tbps. Deploying DDoS protection across the spectrum of attack vectors is no longer a “nice to have,” but a necessity.

In his report, Cooke concludes that “Any DDoS protection product is only part of an overall strategy, not a silver bullet for denial-of-service hazards.” Evaluate your organization and your needs, read more about each solution evaluated in the Radar report, and carefully match the right DDoS solutions to best suit your needs.

Learn More About the Reports: Gigaom Key Criteria for DDoS, and Gigaom Radar for DDoS

The post Defeating Distributed Denial of Service Attacks appeared first on GigaOm.

Continue Reading

Trending