Connect with us

Big Data

GE’s Predix platform bolstered by data, domain expertise

Published

on

BOSTON — CIOs looking for a digital transformation case study will find an ongoing master class at GE. With its Predix platform, which collects and analyzes sensor data from industrial assets such as MRI machines and has been generally available for 18 months, the company is attempting to position itself as the backbone of the industrial internet of things.

But the transformation efforts have been slow to produce results. The company’s earnings are lagging behind industrial competitors — making shareholders uneasy and ultimately leading to the recent departure of CEO Jeff Immelt, digital advocate and proponent of the two-year-old GE Digital.

That was the backdrop for a recent media day event at the company’s temporary headquarters in Boston. Three representatives from GE Digital — Mark Bernardo, vice president of professional services; Mike Varney, senior director of product management; and Jeff Erhardt, vice president of intelligent systems — provided an informal presentation on GE’s Predix platform, the critical role of data and domain expertise for machine learning, and what the future of GE’s young business unit might look like.

Predix platform is key

Immelt was replaced last month by John Flannery, a GE veteran who most recently worked with the company’s healthcare division. One of Flannery’s early tasks as CEO is performing a deep dive into each of GE’s businesses. He plans to complete his audit later this year and present recommendations to investors.

What Flannery’s investigation will mean for the future of the company is yet to be seen. But the representatives from GE Digital said they’ve seen no change in strategy to date and that Immelt’s vision to create the platform for the industrial IoT will likely continue.

In fact, Bernardo, a GE employee for more than 10 years, described reports that GE Digital will need to step up revenue production in 2018 as “normal GE behavior” and not a deviation from strategy.

CIO Jim Fowler expounds on GE’s digital thread and digital twin technologies.

“Our platform, our application investments, our investment in machine learning, our investment in our talent, the reason why domain expertise is important to us is because we need it in order to generate the outcomes our customers need, and to generate the growth and productivity that we need as a business,” he said. “We are as dependent on this strategy as any of our customers.”

With the mention of machine learning, Bernardo is referring, in part, to GE Digital’s 2016 acquisition of Wise.io, a startup out of Berkeley, Calif., that specialized in predicting customer behavior. That may seem like a far cry from industrial assets, but Erhardt, CEO at Wise.io at the time of acquisition, said the key to solving hard problems like predicting customer or machine behavior hinges on a common, underlying data platform that provides a foundation for application development.

“That’s what Salesforce.com has done,” Erhardt said. GE’s Predix platform is built on the same basic model. Erhardt said Wise.io observed from dealings with customers that a data platform is necessary to successfully scale a company based around machine learning, and that it was one of the reasons why being acquired by GE made sense for the startup.

Data is the new oil pipeline

For Wise.io’s part, its job is to make GE applications intelligent. Doing so generally requires computational power and machine learning algorithms — both of which have become commoditized at this point — as well as the increasingly valuable data and domain expertise, according to Erhardt.

“[Data and domain expertise] are at the forefront of both research and how you apply these intelligent techniques, as well as where you can create value,” he said.

He used GE’s intelligent pipeline integrity services products, which rely on the same basic imaging technology packaged in the healthcare business’s products, as an example. “We stick [them] in an oil pipeline and we use [them] to look for defects and weaknesses indicative of that pipeline potentially blowing up,” Erhardt said.

But the technology captures so much data — Erhardt said roughly a terabyte of images — that it can take highly trained experts months to sort out. The machine learning technology, which he defines as “the ability for computers to mimic human decision-making around a data-driven work flow,” relies on past data and decisions to flag problematic areas at super-human speeds.

“The purpose and the idea behind this is to clean up the noise and allow the people to focus on the highest risk, [the] most uncertain areas,” Erhardt said.

The technology doesn’t replace human decision-making outright. Erhardt said his team is spending a good chunk of its time striking the right balance between automation, augmentation and deference. In the latter case, the system defers to domain experts, who may have decades of experience working with complex industrial assets. Domain experts also help GE’s managed service customers prioritize anomalies surfaced by machine learning technology.

Keeping a human in the loop, in other words, is essential. “What’s really important here — and this is different than the consumer space — the cost of being wrong can be very, very high,” Erhardt said.

It’s another reason why machine learning algorithms have to be well-trained, which requires enormous amounts of data. Instead of relying on data generated by a single pipeline integrity product or even a single customer, the Predix platform enables the company to collect and aggregate data across its customer base — and even across its businesses — in a single location. This gives the machine learning tech plenty of training data to learn with and potentially gives GE Digital the raw material to create new revenue streams.

“We’re looking for commonality across these very powerful business cases that exist within our business. What it then gives us the ability to do is to create these derivative products,” Erhardt said. He cited Google’s 2015 acquisition of Waze, an application that helps users avoid traffic jams by using geolocation driver data, as an example of how companies are using data generated by one application to help power other applications. Waze remains a stand-alone application, but the data shared by drivers is now used for city planning purposes.

“The way that we approach this is if you get the core product right — if you can entice your customers to contribute back more data — you not only make that good but you create opportunities you didn’t know about before,” Erhardt said. “That’s what we’re working on.”

PlatoAi. Web3 Reimagined. Data Intelligence Amplified.

Click here to access.

Source: https://searchcio.techtarget.com/news/450426718/GEs-Predix-platform-bolstered-by-data-domain-expertise

Big Data

WHT: A Simpler Version of the fast Fourier Transform (FFT) you should know

Published

on

WHT: A Simpler Version of the fast Fourier Transform (FFT) you should know

The fast Walsh Hadamard transform is a simple and useful algorithm for machine learning that was popular in the 1960s and early 1970s. This useful approach should be more widely appreciated and applied for its efficiency.


By Sean O’Connor, a science and technology author and investigator.

The fast Walsh Hadamard transform (WHT) is a simplified version of the Fast Fourier Transform (FFT.)

The 2-point WHT of the sequence a, b is just the sum and difference of the 2 values:

WHT(a, b) = a+b, a-b. 

It is self-inverse allowing for a fixed constant:

WHT(a+b, a-b) = 2a, 2b 

Due to (a+b) + (a-b) = 2a and (a+b) – (a-b) = 2b.

The constant can be split between the two Walsh Hadamard transforms using a scaling factor of √2 to give a normalized WHTN:

WHTN(a, b) = (a+b)/√2, (a-b)/√2 WHTN((a+b)/√2, (a-b)/√2) = a, b 

That particular constant results in the vector length of a, b being unchanged after transformation since a2+b2 =((a+b)/√2)2+ ((a-b)/√2)2 as you may easily calculate.

The 2-point transform can be extended to longer sequences by sequentially adding and subtracting pairs of similar terms, alike in the pattern of + and – symbols they contain.

To transform a 4-point sequence a, b, c, d first do two 2-point transforms:

WHT(a, b) = a+b, a-b WHT(c, d) = c+d, c-d 

Then add and subtract the alike terms a+b and c+d:

WHT(a+b, c+d) = a+b+c+d, a+b-c-d 

and the alike terms a-b and c-d:

WHT(a-b, c-d) = a-b+c-d, a-b-c+d 

The 4-point transform of a, b, c, d then is

WHT(a, b, c, d) = a+b+c+d,  a+b-c-d, a-b+c-d, a-b-c+d 

When there are no more similar terms to add and subtract, that signals completion (after log2(n) stages, where n is 4 in this case.)  The computational cost of the algorithm is nlog2(n) add/subtract operations, where n, the size of the transform, is restricted to being a positive integer power of 2 in the general case.

If the transform was done using matrix operations, the cost would be much higher (n2 fused multiply-add operations.)

Figure 1.  The 4-point Walsh Hadamard transform calculated in matrix form.

The +1, -1 entries in Figure 1 are presented in a certain natural order which most of the actual algorithms for calculating the WHT result in, which is fortunate since then the matrix is symmetric, orthogonal and self-inverse.

You can also view the +1, -1 patterns of the WHT as waveforms.

Figure 2.  The waveforms of the 8-point WHT presented in natural order.

When you calculate the WHT of a sequence of numbers, you are really just determining how much of each waveform is embedded in the original sequence.  And that is complete and total information with which you can fully reconstruct any sequence from its transform.

The waveforms of the WHT typically correlate strongly with the patterns found in natural data like images, allowing the transform to be used for data compression.

Figure 3.  A 65536-pixel image compressed to 5000 points using a WHT.

In Figure 3, a 65536-pixel image was transformed with a WHT, the 5000 maximum magnitude embeddings were preserved, and then the inverse transform was applied (simply another WHT.)

The central limit theorem (CLT) tells you that adding a large quantity of random numbers results in the Normal distribution with its characteristic bell curve.  The CLT applies equally to sums and differences of a large quantity of random numbers.  As a result, C.M. Rader proposed (in 1969) using the WHT to quickly generate Normally distributed random numbers from conventional uniformly distributed random numbers.  You simply generate a sequence of uniform random numbers, say between –1 and 1, and then transform them using the WHT.

Similarly, you can disrupt the orderly waveform patterns of the WHT by choosing a fixed randomly chosen pattern of sign flips to apply to any input to the transform.  That is equivalent to multiplying the WHT matrix H with a diagonal matrix D of randomly chosen +1, -1 entries giving HD.  The disrupted waveform patterns in HD then fail to correlate with any of the patterns seen in natural data.  As a result, the output of HD has the Normal distribution and is actually a fast Random Projection of the natural data.  Random projections have a wide number of applications in machine learning, such as locality sensitive hashing, compressive sensing, random projection trees, neural network pre and post-processing etc.

References

Walsh (Hadamard) Transform:

Normal Distribution:

Random Projections:

Other Applications:

Related:


PlatoAi. Web3 Reimagined. Data Intelligence Amplified.
Click here to access.

Source: https://www.kdnuggets.com/2021/07/wht-simpler-fast-fourier-transform-fft.html

Continue Reading

Big Data

Must-Know Text Operations in Python before you dive into NLP!

Published

on



Text Operations in Python | Must-Know Text Operations in Python for NLP





















Learn everything about Analytics



PlatoAi. Web3 Reimagined. Data Intelligence Amplified.
Click here to access.

Source: https://www.analyticsvidhya.com/blog/2021/07/must-know-text-operations-in-python-before-you-dive-into-nlp/

Continue Reading

Big Data

Canada’s Rogers Communications beats quarterly revenue estimates

Published

on

(Reuters) -Canada’s Rogers Communications Inc on Wednesday reported second-quarter revenue that beat analysts’ estimates, helped by a pick up in advertisement sales and as its cable business benefited from a pandemic-driven shift to remote work and entertainment.

The requirement of high-speed broadband networks to carry on remote work helped the telecom operator negate the slow recovery from its wireless business.

The return of live sport broadcasting also played a positive role in boosting the Toronto-based telecom operator’s revenue.

The company’s total revenue rose to C$3.58 billion ($2.82 billion) in the quarter ended June 30, compared with analysts’ average estimates of C$3.56 billion, according to IBES data from Refinitiv.

Earlier in March, Rogers said it would buy Shaw Communications Inc for about C$20 billion ($16.02 billion), aiming to double down on its efforts to roll out 5G throughout the country.

Revenue for its cable unit, which includes internet, phone and cloud-based services, rose 5% during the quarter

Quarterly net income rose to C$302 million, or 60 Canadian cents per share, from C$279 million, or 54 Canadian cents, a year earlier.

($1 = 1.2686 Canadian dollars)

(Reporting by Tiyashi Datta in Bengaluru; Editing by Shailesh Kuber)

Image Credit: Reuters

PlatoAi. Web3 Reimagined. Data Intelligence Amplified.
Click here to access.

Source: https://datafloq.com/read/canadas-rogers-communications-beats-quarterly-revenue-estimates/16522

Continue Reading

Big Data

Climate friendly cooling tech firm gets $50 million from Goldman Sachs

Published

on

By Jane Lanhee Lee

(Reuters) – Chemicals used in air conditioning, freezers and refrigeration have long hurt the environment by destroying the ozone layer and polluting water sources, but technology is starting to change the way we keep cool.

Phononic, a startup based in Durham North Carolina using a material called bismuth telluride to make so-called cooling chips, on Wednesday said it raised $50 million from Goldman Sachs Asset Management.

When electricity runs through the chip the current takes heat with it leaving one side of the chip to cool and the other to heat up, said Tony Atti, Phononic co-founder and CEO.

The chips can be as small as a fraction of a fingernail or as big as a fist depending on how much coolants are needed and have been used to create compact freezers for vaccine transportation or for ice-cream at convenience stores like Circle K, he said. A more recent and fast growing use is to prevent overheating in lidars, laser-based sensors in autonomous cars, and optical transceivers for 5G data transmission, said Atti.

“The historical refrigerants that had been used for vapor compression systems, they are both toxic and global warming contributors,” said Atti. While the global warming impact had been reduced, refrigerants still had issues with toxicity and flammability.

Atti said while the bismuth telluride powder itself is toxic, when it is processed into a semiconductor wafer and made into a chip, it is “benign” and can be recycled or disposed as its meets all chip safety and disposal standards.

The cooling chips are manufactured in Phononic’s own factory in Durham and for mass production the company is working with Thailand based Fabrinet. The freezers for vaccines and ice-cream are built in China by contract manufacturers and carry the brands of Phononic’s customers or in some cases are co-branded, he said.

The funding will be used to build out high-volume manufacturing and to expand Phononic’s markets and product line.

Atti declined to share the latest valuation of Phononic but said it was “north of half a billion dollars”. Previous investors include Temasek Holdings and private equity and venture capital firm Oak Investment Partners. 

(Reporting By Jane Lanhee Lee; editing by Richard Pullin)

Image Credit: Reuters

PlatoAi. Web3 Reimagined. Data Intelligence Amplified.
Click here to access.

Source: https://datafloq.com/read/climate-friendly-cooling-tech-firm-gets-50-million-goldman-sachs/16521

Continue Reading
Esports3 days ago

How to reduce lag and increase FPS in Pokémon Unite

Esports4 days ago

Coven skins for Ashe, Evelynn, Ahri, Malphite, Warwick, Cassiopeia revealed for League of Legends

Esports4 days ago

Will New World closed beta progress carry over to the game’s full release?

Aviation5 days ago

And Here’s Yet Another Image Of Russia’s New Fighter Concept That Will Be Officially Unveiled Tomorrow

Esports4 days ago

Can you sprint in New World?

Esports3 days ago

How to add friends and party up in New World

Esports3 days ago

How to claim New World Twitch drops

AR/VR3 days ago

Moth+Flame partners with US Air Force to launch Virtual Reality sexual assault prevention and response training

Esports5 days ago

How to complete FUTTIES Alessandrini’s objectives in FIFA 21 Ultimate Team

Esports3 days ago

Twitch streamer gets banned in New World after milking cow

Esports5 days ago

Everything we know about Seer in Apex Legends

Aerospace5 days ago

Boeing crew capsule mounted on Atlas 5 rocket for unpiloted test flight

Esports5 days ago

What Time Does League of Legends Patch 11.15 Go Live?

Esports5 days ago

Evil Geniuses top laner Impact breaks all-time LCS early-game gold record in win over Dignitas

Blockchain4 days ago

Rothschild Investment Purchases Grayscale Bitcoin and Ethereum Trusts Shares

Blockchain4 days ago

Uniswap (UNI) and AAVE Technical Analysis: What to Expect?

Esports4 days ago

Konami unveils Yu-Gi-Oh! Master Duel, a digital version of the Yu-Gi-Oh! TCG and OCG formats

Blockchain3 days ago

BNY Mellon Joins State Street Into Crypto Trading, Backs Pure Digital Trading Platform

Esports3 days ago

How to change or join a new world in New World

Esports4 days ago

Team BDS adds GatsH to VALORANT roster as sixth man before EU Stage 3 Challengers 2

Trending