Page 21234..10..»

Archive for the ‘Machine Learning’ Category


Posted: September 20, 2020 at 10:56 pm

without comments

A joint venture has seen the implementation of machine learning at HHLAs Container Terminal Burchardkai to optimise import container yard positioning and reduce re-handling moves.

The elimination of costly re-handling moves of import containers has recently been the focus of a joint project between container terminal operator HHLA, its affi liate Hamburg Port Consulting (HPC) and INFORM the Artificial Intelligence (AI) systems supplier. Machine learning sits at the heart of the system.

Dwell time is the unit of time used to measure the period in which a container remains in a container terminal with this typically running from its arrival off a vessel until leaving the terminal via truck, rail or another vessel.

For import containers there is often no specific information available on the pick-up time when selecting a storage slot in the container stack. This can lead to an inefficient container storage location in the yard generating, in turn, the requirement for additional shuffle moves that require extra resources including maintenance and energy consumption.

To mitigate this operational inefficiency, the project partners - HHLA, HPC and INFORM - have recently run a pilot project at HHLAs Container Terminal Burchardkai (CTB) focused on machine learning technology with this applied in order to predict individual import container dwell times and thereby reduce costly re-handling/shuffle moves.

As a specialist in IT software integration and terminal operations, HPC employed the deep learning approach to identify hidden patterns from historical data of container moves at HHLA CTB. This was undertaken over a period of two years and with the acquired information processed into high quality data sets. Assessed by the Syncrotess Machine Learning Module from INFORM and validated by the HPC simulation tool, the results show a significant reduction of shuffle moves resulting in a reduced truck turn time.


Dr. Alexis Pangalos, Partner at HPC discussing the project highlights notes: It was a productive implementation of INFORMs Artificial Intelligence (AI) solution for the choice of container storage positions at CTB. The Machine Learning (ML) Module was trained with data from CTBs container handling operations and the outcome from this is a system tailor-made for HHLAs operations.

HPC together with INFORM have integrated the Syncrotess ML Module into the slot allocation algorithms already running within CTBs terminal control system, ITS.


INFORMs AI solution predicts the dwell time (i.e., the time period the container is expected to be stored in the yard) and the outbound mode of transport (e.g., rail, truck, vessel) both of which are crucial criteria for selecting an optimised container storage location within the yard. A location that avoids unnecessary re-handling.

Utilising machine learning and AI and integrating these technologies into existing IT infrastructure are the success factors for reaching the next level of optimisations, says Jens Hansen, Executive Board Member responsible for IT at HHLA. A detailed analysis, and a smooth interconnectivity between all different systems, enable the value of improved safety while reducing costs and greenhouse gas emissions, he underlines.


Data availability and data processing are key elements when it comes to utilising AI technology, says Pangalos. It requires a detailed domain knowledge of terminal operations to unlock greater productivity of the terminal equipment and connected processes.

The implementation is based on a machine learning assessment INFORM undertook in 2018 whereby it set out to determine if they could improve optimisation and operational outcomes using INFORMs broader ML algorithms developed for use in other industries such as finance and aviation.

As of 2019, system results indicated a prediction accuracy of 26% for dwell time predictions and 33% for outbound mode of transport predictions.

Dr. Eva Savelsberg, Senior Vice President of INFORMs Logistic Division notes: AI and machine learning allows us to leverage data from our past performance to inform us about how best to approach our future operations our ML Module gives our Operations Research based algorithms the best footing for making complex decisions about what to do in the future.

INFORMs Machine Learning Module allows CTB to leverage insights generated from algorithms that continuously learn from historical data."

Further Information: Matthew Wittemeier

Continued here:


Written by admin

September 20th, 2020 at 10:56 pm

Posted in Machine Learning

What is ‘custom machine learning’ and why is it important for programmatic optimisation? – The Drum

Posted: at 10:56 pm

without comments

Wayne Blodwell, founder and chief exec of The Programmatic Advisory & The Programmatic University, battles through the buzzwords to explain why custom machine learning can help you unlock differentiation and regain a competitive edge.

Back in the day, simply having programmatic on plan was enough to give you a competitive advantage and no one asked any questions. But as programmatic has grown, and matured (84.5% of US digital display spend is due to be bought programmatically in 2020, the UK is on track for 92.5%), whats next to gain advantage in an increasingly competitive landscape?

Machine Learning


The use and development of computer systems that are able to learn and adapt without following explicit instructions, by using algorithms and statistical models to analyse and draw inferences from patterns in data.

(Oxford Dictionary, 2020)

Youve probably head of machine learning as it exists in many Demand Side Platforms in the form of automated bidding. Automated bidding functionality does not require a manual CPM bid input nor any further bid adjustments instead, bids are automated and adjusted based on machine learning. Automated bids work from goal inputs, eg achieve a CPA of x or simply maximise conversions, and these inputs steer the machine learning to prioritise certain needs within the campaign. This tool is immensely helpful in taking the guesswork out of bids and the need for continual bid intervention.

These are what would be considered off-the-shelf algorithms, as all buyers within the DSP have access to the same tool. There is a heavy reliance on this automation for buying, with many even forgoing traditional optimisations for fear of disrupting the learnings and holding it back but how do we know this approach is truly maximising our results?

Well, we dont. What we do know is that this machine learning will be reasonably generic to suit the broad range of buyers that are activating in the platforms. And more often than not, the functionality is limited to a single success metric, provided with little context, which can isolate campaign KPIs away from their true overarching business objectives.

Custom machine learning

Instead of using out of the box solutions, possibly the same as your direct competitors, custom machine learning is the next logical step to unlock differentiation and regain an edge. Custom machine learning is simply machine learning that is tailored towards specific needs and events.

Off-the-self algorithms are owned by the DSPs; however, custom machine learning is owned by the buyer. The opportunity for application is growing, with leading DSPs opening their APIs and consoles to allow for custom logic to be built on top of existing infrastructure. Third party machine learning partners are also available, such as Scibids, MIQ & 59A, which will develop custom logic and add a layer onto the DSPs to act as a virtual trader, building out granular strategies and approaches.

With this ownership and customisation, buyers can factor in custom metrics such as viewability measurement and feed in their first party data to align their buying and success metrics with specific business goals.

This level of automation not only provides a competitive edge in terms of correctly valuing inventory and prioritisation, but the transparency of the process allows trust to rightfully be placed with automation.

Custom considerations

For custom machine learning to be effective, there are a handful of fundamental requirements which will help determine whether this approach is relevant for your campaigns. Its important to have conversations surrounding minimum event thresholds and campaign size with providers, to understand how much value you stand to gain from this path.

Furthermore, a custom approach will not fix a poor campaign. Custom machine learning is intended to take a well-structured and well-managed campaign and maximise its potential. Data needs to be inline for it to be adequately ingested and for real insight and benefit to be gained. Custom machine learning cannot simply be left to fend for itself; it may lighten the regular day to day load of a trader, but it needs to be maintained and closely monitored for maximum impact.

While custom machine learning brings numerous benefits to the table transparency, flexibility, goal alignment its not without upkeep and workflow disruption. Levels of operational commitment may differ depending on the vendors selected to facilitate this customisation and their functionality, but generally buyers must be willing to adapt to maximise the potential that custom machine learning holds.

Find out more on machine learning in a session The Programmatic University are hosting alongside Scibids on The Future Of Campaign Optimisation on 17 September. Sign up here.

Read more:

What is 'custom machine learning' and why is it important for programmatic optimisation? - The Drum

Written by admin

September 20th, 2020 at 10:56 pm

Posted in Machine Learning

How Machine Learning is Set to Transform the Online Gaming Community – –

Posted: at 10:56 pm

without comments

We often equate machine learning to fictional scenarios such as those presented in films including the Terminator franchise and 2001: A Space Odyssey. While these are all entertaining stories, the fact of the matter is that this type of artificial intelligence is not nearly as threatening. On the contrary, it has helped to dramatically enhance the overall user experience (UX) and to streamline many online functions (such as common search results) that we take for granted. Machine learning is also making its presence known within the digital gaming community. Without becoming overly technical, what transformations can we expect to witness and how will these impact the experience of the average gaming enthusiast?

Although games such as Pong and Super Mario Bros. were entertaining for their time, they were also quite predictable. This is why so many users have uploaded speed runs onto websites such as YouTube. However, what if a game actually learned from your previous actions? It is obvious that the platform itself would be much more challenging. This concept is now becoming a reality.

Machine learning can also apply to numerous scenarios. It may be used to provide a greater sense of realism with interacting with a role-playing game. It could be employed to offer speech recognition and to recognise voice commands. Machine learning may also be implemented to create more realistic non-playable characters (NPCs).

Whether referring to fast-paced MMORPGs to traditional forms of entertainment including slot games offered by websites such as, there is no doubt that machine learning will soon make its presence known.

We can clearly see that the technical benefits associated with machine learning will certainly be leveraged by game developers. However, it is just as important to mention that this very same technology will have a pronounced impact upon the players themselves. This is largely due to how games can be personalised based around the needs of the player.

We are not only referring to common options such as the ability to modify avatars and skins in this case. Instead, games are evolving to the point that they will base their recommendations off of the behaviours of the players themselves. For example, a plot may change as a result of how a player interacts with other characters. The difficulty of a specific level may be automatically adjusted in accordance with the skill of the player. As machine learning and AI both have the ability to model extremely complex systems, the sheer attention to graphical detail within the games (such as character features and backgrounds) will also become vastly enhanced.

We can see that the future of gaming looks extremely bright thanks to the presence of machine learning. While such systems might appear to have little impact upon traditional platforms such as solitaire, there is no doubt that they will still be felt across numerous other genres. So, get ready for a truly amazing experience in the months and years to come!

See the rest here:

How Machine Learning is Set to Transform the Online Gaming Community - -

Written by admin

September 20th, 2020 at 10:56 pm

Posted in Machine Learning

Current and future regulatory landscape for AI and machine learning in the investment management sector – Lexology

Posted: at 10:56 pm

without comments

On Tuesday this week, Mark Lewis, senior consultant in IT, fintech and outsourcing at Macfarlanes, took part in an event hosted by The Investment Association covering some of the use cases, successes and challenges faced when implementing AI and machine learning (AIML) in the investment management industry.

Mark led the conversation on the current regulatory landscape for AIML and on the future direction of travel for the regulation of AIML in the investment management sector. He identified several challenges posed by the current regulatory framework, including those caused by the lack of a standard definition of AI generally and for regulatory purposes. This creates the risk of a fragmented regulatory landscape (an expression used recently by the World Federation of Exchanges in the context of lack of a standard taxonomy for fintech globally) as different regulators tend to use different definitions of AIML. This results in the risk of over- or under-regulating AIML and is thought to be inhibiting firms adopting new AI systems. While the UK Financial Conduct Authority (FCA) and the Bank of England seem to have settled, at least for now, on a working definition of AI as the use of a machine to perform tasks normally requiring human intelligence, and of ML as a subset of AI where a machine teaches itself to perform tasks without being explicitly programmed these working definitions are too generic to be of serious practical use in approaching regulation.

The current raft of legislation and other regulation that can apply to AI systems is uncertain, vast and complex, particularly within the scope of regulated financial services. Part of the challenge is that, for now, there is very little specific regulation directly applicable to AIML (exceptions include GDPR and, for algorithmic high-frequency trading, MiFID II). The lack of understanding of new AIML systems, combined with an uncertain and complex regulatory environment, also has an impact internally within businesses as they attempt to implement these systems. Those responsible for compliance are reluctant to engage where sufficient evidence is not available on how the systems will operate and how great the compliance burden will be. Improvements in explanations from technologists may go some way to assisting in this area. Overall, this means that regulated firms are concerned that their current systems and governance processes for technology, digitisation and related services deployments remain fit-for-purpose when extended to AIML. They are seeking reassurance from their regulators that this is the case. Firms are also looking for informal, discretionary regulatory advice on specific AIML concerns, such as required disclosures to customers about the use of chatbots.

Aside from the sheer volume of regulation that could apply to AIML development and deployment, there is complexity in the sources of regulation. For example, firms must also have regard to AIML ethics and ethical standards and policies. In this context, Mark noted that, this year, the FCA and The Alan Turing Institute launched a collaboration on transparency and explainability of AI in the UK financial services sector, which will lead to the publication of ethical standards and expectations for firms deploying AIML. He also referred to the role of the UK governments Centre for Data Ethics and Innovation (CDEI) in the UKs regulatory framework for AI and, in particular to the CDEIs AI Barometer Report (June 2020), which has clearly identified several key areas that will most likely require regulatory attention, and some with significant urgency. These include:

In the absence of significant guidance, Mark provided a practical, 10-point, governance plan to assist firms in developing and deploying AI in the current regulatory environment, which is set out below. He highlighted the importance of firms keeping watch on regulatory developments, including what regulators and their representatives say about AI, as this may provide an indication of direction in the absence of formal advice. He also advised that firms ignore ethics considerations at their peril, as these will be central to any regulation going forward. In particular, for the reasons given above, he advised keeping up to date with reports from the CDEI. Other topics discussed in the session included lessons learnt for best practice in the fintech industry and how AI has been used to solve business challenges in financial markets.

See the original post here:

Current and future regulatory landscape for AI and machine learning in the investment management sector - Lexology

Written by admin

September 20th, 2020 at 10:56 pm

Posted in Machine Learning

Global Machine Learning Courses Market Research Report 2015-2027 of Major Types, Applications and Competitive Vendors in Top Regions and Countries -…

Posted: at 10:56 pm

without comments

Strategic growth, latest insights, developmental trends in Global & Regional Machine Learning Courses Market with post-pandemic situations are reflected in this study. End to end Industry analysis from the definition, product specifications, demand till forecast prospects are presented. The complete industry developmental factors, historical performance from 2015-2027 is stated. The market size estimation, Machine Learning Courses maturity analysis, risk analysis, and competitive edge is offered. The segmental market view by types of products, applications, end-users, and top vendors is stated. Market drivers, restraints, opportunities in Machine Learning Courses industry with the innovative and strategic approach is offered. Machine Learning Courses product demand across regions like North America, Europe, Asia-Pacific, South and Central America, Middle East, and Africa is analyzed. The emerging segments, CAGR, revenue accumulation, feasibility check is specified.

Know more about this report or browse reports of your interest here:

COVID-19 has greatly impacted different Machine Learning Courses segments causing disruptions in the supply chain, timely product deliveries, production processes, and more. Post pandemic era the Machine Learning Courses industry will emerge with completely new norms, plans and policies, and development aspects. There will be new risk factors involved along with sustainable business plans, production processes, and more. All these factors are deeply analyzed by Reports Check's domain expert analysts for offering quality inputs and opinions.

Check out the complete table of contents, segmental view of this industry research report:

The qualitative and quantitative information is formulated in Machine Learning Courses report. Region-wise or country-wise reports are exclusively available on clients' demand with Reports Check. The market size estimation, Machine Learning Courses industry's competition, production capacity is evaluated. Also, import-export details, pricing analysis, upstream raw material suppliers, and downstream buyers analysis is conducted.

Receive complete insightful information with past, present and forecast situations of Global Machine Learning Courses Market and Post-Pandemic Status. Our expert analyst team is closely monitoring the industry prospects and revenue accumulation. The report will answer all your queries as well as you can make a custom request with free sample report.

A full-fledged, comprehensive research technique is used to derive Machine Learning Courses market's quantitative information. The gross margin, Machine Learning Courses sales ratio, revenue estimates, profits, and consumer analysis is provided. The complete global Machine Learning Courses market size, regional, country-level market size, & segmentation-wise market growth and sales analysis are provided. Value chain optimization, trade policies, regulations, opportunity analysis map, & marketplace expansion, and technological innovations are stated. The study sheds light on the sales growth of regional and country-level Machine Learning Courses market.

The company overview, total revenue, Machine Learning Courses financials, SWOT analysis, and product launch events are specified. We offer competitor analysis under the competitive landscape section for every competitor separately. The report scope section provides in-depth analysis of overall growth, leading companies with their successful Machine Learning Courses marketing strategies, market contribution, recent developments, and historic and present status.

Segment 1: Describes Machine Learning Courses market overview with definition, classification, product picture, Machine Learning Courses specifications

Segment 2: Machine Learning Courses opportunity map, market driving forces, restraints, and risk analysis

Segment 3:Competitive landscape view, sales, revenue, gross margin, pricing analysis, and global market share analysis

Segment 4:Machine Learning Courses Industry fragments by key types, applications, top regions, countries, top companies/manufacturers and end-users

Segment 5:Regional level growth, sales, revenue, gross margin from 2015-2020

Segment 6,7,8:Country-level sales, revenue, growth, market share from 2015-2020

Segment 9:Market sales, size, and share by each product type, application, and regional demand with production and Machine Learning Courses volume analysis

Segment 10:Machine Learning Courses Forecast prospects situations with estimates revenue generation, share, growth rate, sales, demand, import-export, and more

Segment 11 & 12:Machine Learning Courses sales and marketing channels, distributor analysis, customers, research findings, conclusion, and analysts views and opinions

Click to know more about our company and service offerings:

An efficient research technique with verified and reliable data sources is what makes us stand out of the crowd. Excellent business approach, diverse clientele, in-depth competitor analysis, and efficient planning strategy is what makes us stand out of the crowd. We cater to different factors like technological innovations, economic developments, R&D, and mergers and acquisitions are specified. Credible business tactics and extensive research is the key to our business which helps our clients in profitable business plans.

Contact Us:

Olivia Martin

Email: [emailprotected]

Phone: +1(831)6793317

See the rest here:

Global Machine Learning Courses Market Research Report 2015-2027 of Major Types, Applications and Competitive Vendors in Top Regions and Countries -...

Written by admin

September 20th, 2020 at 10:56 pm

Posted in Machine Learning

When AI in healthcare goes wrong, who is responsible? – Quartz

Posted: at 10:56 pm

without comments

Artificial intelligence can be used to diagnose cancer, predict suicide, and assist in surgery. In all these cases, studies suggest AI outperforms human doctors in set tasks. But when something does go wrong, who is responsible?

Theres no easy answer, says Patrick Lin, director of Ethics and Emerging Sciences Group at California Polytechnic State University. At any point in the process of implementing AI in healthcare, from design to data and delivery, errors are possible. This is a big mess, says Lin. Its not clear who would be responsible because the details of why an error or accident happens matters. That event could happen anywhere along the value chain.

Design includes creation of both hardware and software, plus testing the product. Data encompasses the mass of problems that can occur when machine learning is trained on biased data, while deployment involves how the product is used in practice. AI applications in healthcare often involve robots working with humans, which further blurs the line of responsibility.

Responsibility can be divided according to where and how the AI system failed, says Wendall Wallace, lecturer at Yale Universitys Interdisciplinary Center for Bioethics and the author of several books on robot ethics. If the system fails to perform as designed or does something idiosyncratic, that probably goes back to the corporation that marketed the device, he says. If it hasnt failed, if its being misused in the hospital context, liability would fall on who authorized that usage.

Surgical Inc., the company behind the Da Vinci Surgical system, has settled thousands of lawsuits over the past decade. Da Vinci robots always work in conjunction with a human surgeon, but the company has faced allegations of clear error, including machines burning patients and broken parts of machines falling into patients.

Some cases, though, are less clear-cut. If diagnostic AI trained on data that over-represents white patients then misdiagnoses a Black patient, its unclear whether the culprit is the machine-learning company, those who collected the biased data, or the doctor who chose to listen to the recommendation. If an AI program is a black box, it will make predictions and decisions as humans do, but without being able to communicate its reasons for doing so, writes attorney Yavar Bathaee in a paper outlining why the legal principles that apply to humans dont necessarily work for AI. This also means that little can be inferred about the intent or conduct of the humans that created or deployed the AI, since even they may not be able to foresee what solutions the AI will reach or what decisions it will make.

The difficulty in pinning the blame on machines lies in the impenetrability of the AI decision-making process, according to a paper on tort liability and AI published in the AMA Journal of Ethics last year. For example, if the designers of AI cannot foresee how it will act after it is released in the world, how can they be held tortiously liable?, write the authors. And if the legal system absolves designers from liability because AI actions are unforeseeable, then injured patients may be left with fewer opportunities for redress.

AI, as with all technology, often works very differently in the lab than in a real-world setting. Earlier this year, researchers from Google Health found that a deep-learning system capable of identifying symptoms of diabetic retinopathy with 90% accuracy in the lab caused considerable delays and frustrations when deployed in real life.

Despite the complexities, clear responsibility is essential for artificial intelligence in healthcare, both because individual patients deserve accountability, and because lack of responsibility allows mistakes to flourish. If its unclear whos responsible, that creates a gap, it could be no one is responsible, says Lin. If thats the case, theres no incentive to fix the problem. One potential response, suggested by Georgetown legal scholar David Vladeck, is to hold everyone involved in the use and implementation of the AI system accountable.

AI and healthcare often work well together, with artificial intelligence augmenting the decisions made by human professionals. Even as AI develops, these systems arent expected to replace nurses or automate human doctors entirely. But as AI improves, it gets harder for humans to go against machines decisions. If a robot is right 99% of the time, then a doctor could face serious liability if they make a different choice. Its a lot easier for doctors to go along with what that robot says, says Lin.

Ultimately, this means humans are ceding some authority to robots. There are many instances where AI outperforms humans, and so doctors should defer to machine learning. But patient wariness of AI in healthcare is still justified when theres no clear accountability for mistakes. Medicine is still evolving. Its part art and part science, says Lin. You need both technology and humans to respond effectively.

Read the original:

When AI in healthcare goes wrong, who is responsible? - Quartz

Written by admin

September 20th, 2020 at 10:56 pm

Posted in Machine Learning

Is Wide-Spread Use of AI & Machine Intelligence in Manufacturing Still Years Away? – Automation World

Posted: at 10:56 pm

without comments

According to a new report by PMMI Business Intelligence, artificial intelligence (AI) and machine learning is the area of automation technology with the greatest capacity for expansion. This technology can optimize individual processes and functions of the operation; manage production and maintenance schedules; and, expand and improve the functionality of existing technology such as vision inspection.

While AI is typically aimed at improving operation-wide efficiency, machine learning is directed more toward the actions of individual machines; learning during operation, identifying inefficiencies in areas such as rotation and movement, and then adjusting processes to correct for inefficiencies.

The advantages to be gained through the use of AI and machine learning are significant. One study released by Accenture and Frontier Economics found that by 2035, AI-empowered technology could increase labor productivity by up to 40%, creating an additional $3.8 trillion in direct value added (DVA) to the manufacturing sector.

See it Live at PACK EXPO Connects Nov. 9-13: End-of-Line Automation without Capital Expenditure, by Pearson Packaging Systems. Preview the Showroom Here.

However, only 1% of all manufacturers, both large and small, are currently utilizing some form of AI or machine learning in their operations. Most manufacturers interviewed said that they are trying to gain a better understanding of how to utilize this technology in their operations, and 45% of leading CPGs interviewed predict they will incorporate AI and/or machine learning within ten years.

A plant manager at a private label SME reiterates AI technology is still being explored, stating: We are only now talking about how to use AI and predict it will impact nearly half of our lines in the next 10 years.

While CPGs forecast that machine learning will gain momentum in the next decade, the near-future applications are likely to come in vision and inspection systems. Manufacturers can utilize both AI and machine learning in tandem, such as deploying sensors to key areas of the operation to gather continuous, real-time data on efficiency, which can then be analyzed by an AI program to identify potential tweaks and adjustments to improve the overall process.

See it Live at PACK EXPO Connects Nov. 9-13: Reduce costs and improve product quality in adhesive application of primary packaging, by Robatech USA Inc. Preview the Showroom Here.

And, the report states, that while these may appear to be expensive investments best left for the future, these technologies are increasingly affordable and offer solutions that can bring measurable efficiencies to smart manufacturing. In the days of COVID-19, gains to labor productivity and operational efficiency may be even more timely.

To access this FREE report and learn more about automation in operations, download below.

Source: PMMI Business Intelligence, Automation Timeline: The Drive Toward 4.0 Connectivity in Packaging and Processing

PACK EXPO Connects November 9-13. Now more than ever, packaging and processing professionals need solutions for a rapidly changing world, and the power of the PACK EXPO brand delivers the decision makers you need to reach. Attendeeregistrationis open now.

Read more here:

Is Wide-Spread Use of AI & Machine Intelligence in Manufacturing Still Years Away? - Automation World

Written by admin

September 20th, 2020 at 10:56 pm

Posted in Machine Learning

How do we know AI is ready to be in the wild? Maybe a critic is needed – ZDNet

Posted: at 10:56 pm

without comments

Mischief can happen when AI is let loose in the world, just like any technology. The examples of AI gone wrong are numerous, the most vivid in recent memory being the disastrously bad performance of Amazon's facial recognition technology, Rekognition, which had a propensity to erroneously match members of some ethnic groups with criminal mugshots to a disproportionate extent.

Given the risk, how can society know if a technology has been adequately refined to a level where it is safe to deploy?

"This is a really good question, and one we are actively working on, "Sergey Levine, assistant professor with the University of California at Berkeley's department of electrical engineering and computer science, told ZDNet by email this week.

Levine and colleagues have been working on an approach to machine learning where the decisions of a software program are subjected to a critique by another algorithm within the same program that acts adversarially. The approach is known as conservative Q-Learning, and it was described in a paper posted on the arXiv preprint server last month.

ZDNet reached out to Levine this week after he posted an essay on Medium describing the problem of how to safely train AI systems to make real-world decisions.

Levine has spent years at Berkeley's robotic artificial intelligence and learning lab developing AI software that to direct how a robotic arm moves within carefully designed experiments-- carefully designed because you don't want something to get out of control when a robotic arm can do actual, physical damage.

Robotics often relies on a form of machine learning called reinforcement learning. Reinforcement learning algorithms are trained by testing the effect of decisions and continually revising a policy of action depending on how well the action affects the state of affairs.

But there's the danger: Do you want a self-driving car to be learning on the road, in real traffic?

In his Medium post, Levine proposes developing "offline" versions of RL. In the offline world, RL could be trained using vast amounts of data, like any conventional supervised learning AI system, to refine the system before it is ever sent out into the world to make decisions.

Also: A Berkeley mash-up of AI approaches promises continuous learning

"An autonomous vehicle could be trained on millions of videos depicting real-world driving," he writes. "An HVAC controller could be trained using logged data from every single building in which that HVAC system was ever deployed."

To boost the value of reinforcement learning, Levine proposes moving from the strictly "online" scenario, exemplified by the diagram on the right, to an "offline" period of training, whereby algorithms are input with masses of labeled data more like traditional supervised machine learning.

Levine uses the analogy of childhood development. Children receive many more signals from the environment than just the immediate results of actions.

"In the first few years of your life, your brain processed a broad array of sights, sounds, smells, and motor commands that rival the size and diversity of the largest datasets used in machine learning," Levine writes.

Which comes back to the original question, to wit, after all that offline development, how does one know when an RL program is sufficiently refined to go "online," to be used in the real world?

That's where conservative Q-learning comes in. Conservative Q-learning builds on the widely studied Q-learning, which is itself a form of reinforcement learning. The idea is to "provide theoretical guarantees on the performance of policies learned via offline RL," Levine explained to ZDNet. Those guarantees will block the RL system from carrying out bad decisions.

Imagine you had a long, long history kept in persistent memory of what actions are good actions that prevent chaos. And imagine your AI algorithm had to develop decisions that didn't violate that long collective memory.

"This seems like a promising path for us toward methods with safety and reliability guarantees in offline RL," says UC Berkeley assistant professor Sergey Levine, of the work he and colleagues are doing with "conservative Q-learning."

In a typical RL system, a value function is computed based on how much a certain choice of action will contribute to reaching a goal. That informs a policy of actions.

In the conservative version, the value function places a higher value on that past data in persistent memory about what should be done. In technical terms, everything a policy wants to do is discounted, so that there's an extra burden of proof to say that the policy has achieved its optimal state.

A struggle ensues, Levine told ZDNet, making an analogy to generative adversarial networks, or GANs, a type of machine learning.

"The value function (critic) 'fights' the policy (actor), trying to assign the actor low values, but assign the data high values." The interplay of the two functions makes the critic better and better at vetoing bad choices. "The actor tries to maximize the critic," is how Levine puts it.

Through the struggle, a consensus emerges within the program. "The result is that the actor only does those things for which the critic 'can't deny' that they are good (because there is too much data that supports the goodness of those actions)."

Also: MIT finally gives a name to the sum of all AI fears

There are still some major areas that need refinement, Levine told ZDNet. The program at the moment has some hyperparameters that have to be designed by hand rather than being arrived at from the data, he noted.

"But so far this seems like a promising path for us toward methods with safety and reliability guarantees in offline RL," said Levine.

In fact, conservative Q-learning suggests there are ways to incorporate practical considerations into the design of AI from the start, rather than waiting till after such systems are built and deployed.

Also: To Catch a Fake: Machine learning sniffs out its own machine-written propaganda

The fact that it is Levine carrying out this inquiry should give the approach of conservative Q-learning added significance. With a firm grounding in real-world applications of robotics, Levine and his team are in a position to validate the actor-critic in direct experiments.

Indeed, the conservative Q-Learning paper, which is lead-authored by Aviral Kumar of Berkeley, and was done with the collaboration of Google Brain, contains numerous examples of robotics tests in which the approach showed improvements over other kinds of offline RL.

There is also a blog post authored by Google if you want to learn more about the effort.

Of course, any system that relies on amassed data offline for its development will be relying on the integrity of that data. A successful critique of the kind Levine envisions will necessarily involve broader questions about where that data comes from, and what parts of it represent good decisions.

Some aspects of what is good and bad may be a discussion society has to have that cannot be automated.

Excerpt from:

How do we know AI is ready to be in the wild? Maybe a critic is needed - ZDNet

Written by admin

September 20th, 2020 at 10:56 pm

Posted in Machine Learning

Solving the crux behind Apple’s Silicon Strategy – Medium

Posted: at 10:56 pm

without comments

In its latest keynote address headed by CEO Tim Cook, Apple its new A14 bionic chip, a 5 nm ARM based chipset.

This System on a Chip (SoC) from Apple is expected to power iPhone 12 and iPad Air (2020) models. The chipset integrates around 11.8 billion transistors.

For over a decade, Apples world-class silicon design team has been building and refining Apple SoCs. Using these designs Apple has been able to develop the latest iPhone, iPad and Apple Watch that are the industry leaders in terms of class and performance. In June of 2020, Apple announced that it will transition the Mac to its custom silicon to offer better technological performance.

Now, Apple Silicon is basically a processor made in-house akin to what is powering the iPhone and iPad family of devices. This ARM move will result in ditching their reliance on Intel chipsets for Future Macs. This transition to silicon will also establish a common architecture across all Apple products, making it far easier for developers to write and optimize their apps for the entire ecosystem. In fact, developers can now start focusing on updating their applications to take advantage of the enhanced capabilities of the Apple silicon.

Along with this Apple also introduced mac0S Big Sur earlier this year, which will be the next major macOS release (version 11.0) and includes technologies that will facilitate a smooth transition to the Apple silicon experience. This will be the first time where developers will be able to make their iOS and iPad OS apps available on the Mac without modifications. The Apple silicon powered Macs will offer industry leading performance per watt and higher performance GPUs. To help developers get accustomed to the new transition, Apple is also launching the Universal App QuickStart Program to guide developers through the entire transition.

Apple plans to ship the new Mac by the end of the year and complete the transition in about two years. This being said Apple will continue to release new versions for Intel-based Mac for years to come.

Apple has been explicit about how serious they are about machine learning-based SoC. Apple A14 includes second-generation machine learning accelerators in the CPU for 10 times faster machine learning calculations. The combination of the new Neural Engine, machine learning accelerators, advanced power management, unified memory architecture and the Apple high-performance GPU enables powerful on-device experiences for image recognition, natural language learning, analysing motion, and maybe a machine learning enabled GPS!

According to a recent patent application by Apple , they have been working on a technology that implements a system for estimating the device location based on a global positioning system consisting of a Global Navigation Satellite System (GNSS) satellite, and receives a set of parameters associated with the estimated position. The processor is further configured to apply the set of parameters and the estimated position to a machine learning model that has been trained on a position relative to the satellite. The estimated position and output of the model is then provided to a Kalman filter for more accurate location.

This technology may be significantly better than what a mobile device alone can perform in most non-aided mode(s) of operation. Apples patent to improve GPS in the upcoming 5G era might give them an advantage over existing resources.

Apples move to its own ARM chips comes just as the company unveils macOS version 11.0 (Big Sur). That means ARM based Mac computers will continue to run macOS instead of switching to iOS 14, similar to the approach taken with existing Windows laptops that use Qualcomm ARM based processors. Apple apparently has its hardware and software team working together, given that they have found a way for all their applications functioning seamless from day one of the launch, through Rosetta 2 acting as an emulator and a translator that will allow Intel-made apps to run on Silicon-powered devices.

Moreover, the Apple ecosystem acts as the catalyst for innovation in the company and is not limited to the hardware and software products, but also around its services.

Putting a foot forward in that direction is the Apple One Subscription.

Apple with its calm dignity, diligent market study and unflinching courage to innovate has taken its own time to come up with their strategic silicon move. Apple stayed focused on its long term goals instead of following the hype, trends and gimmicks set out by its competitors to gain customer attention. This ability to think differently is a driving force behind their success.

And owing to the current state of affairs Apple has played it relatively safe this year, sticking to their core offerings. We can expect an exciting iPhone, iMac and MacOS launch later this year.

Lets gear up for another round of innovation sponsored by Apple.

Continued here:

Solving the crux behind Apple's Silicon Strategy - Medium

Written by admin

September 20th, 2020 at 10:56 pm

Posted in Machine Learning

Boost Your Animation To 60 FPS Using AI – Hackaday

Posted: at 10:56 pm

without comments

The uses of artificial intelligence and machine learning continue to expand, with one of the more recent implementations being video processing. A new method can fill in frames to smooth out the appearance of the video, which [LegoEddy] was able to use this in one of his animated LEGO movies with some astonishing results.

His original animation of LEGO figures and sets was created at 15 frames per second. As an animator, he notes that its orders of magnitude more difficult to get more frames than this with traditional methods, at least in his studio. This is where the artificial intelligence comes in. The program is able to interpolate between frames and create more frames to fill the spaces between the original. This allowed [LegoEddy] to increase his frame rate from 15 fps to 60 fps without having to actually create the additional frames.

While weve seen AI create art before, the improvement on traditionally produced video is a dramatic advancement. Especially since the AI is aware of depth and preserves information about the distance of objects from the camera. The software is also free, runs on any computer with an appropriate graphics card, and is available on GitHub.

Thanks to [BaldPower] for the tip!

See the original post:

Boost Your Animation To 60 FPS Using AI - Hackaday

Written by admin

September 20th, 2020 at 10:56 pm

Posted in Machine Learning

Page 21234..10..»