With Neither AI nor Big Data, Is There Any Clue to See through the Blockchain Stuff?
At present, we are in the revolution of artificial intelligence, an age with a sprouting number of new researches and applications.
Advances in artificial intelligence (such as deep learning) have changed everything, from intelligent recommendations on eBusiness platforms to voice assistants in tour-guide map gadgets.
It seems that quite a few tasks that used to rely on people are now accomplished only by machines.
Not long ago, AI seemed just a part of science fiction.
Today, AI glows ubiquitously.
However, the emergence of this new technology is also tempting people to harbour many unrealistic expectations, such as odd publicities over machine learning.
For those standing aside, it is still worthwhile to know how people overestimated the power of machine learning and what may boost machine learning to go further. As for my attitude, a fully-activated machine learning eventually needs embedding in blockchain technology.
The term “machine learning” comes from the basic theory of Arthur Samuel. In 1959, he figured it out in his ML fundamental theory that machines can learn to do certain things by identifying and analysing certain behaviours. In such a condition, programmers no longer have to program individually for each micro-specific task.
In essence, it is to let the machine learn from big data.
At present, a growing volume of investment is pouring into data centres: data centres, performing machine learning tasks, ground their computing methods on traditional CPUs. A typical CPU can hold 6 ~ 14 kernels and can run between 12 ~ 28 command threads.
Typically, these threads are circumscribed in a single block of data. Therefore, building more CPU data centres in such an approach is not proportionately and sufficiently meeting a growing need from AI.
However, there is another type of computation. It can better response a growing demand for AI’s computing power: GPU-based computation. A typical GPU can uphold 2, 000, 000 cores, and each thread can sustain not less than 100 command threads.
Typically, these threads will run on about 30 blocks of information at the same time. With such a kind of computational power, during processing and allocating, the GPU-based computation method can accelerate the whole process and deduce the energy consumption. Therefore, it indeed caters to for machine learning tasks.
Blockchain or distributed ledger technology (DLT) can provide the computing resources needed by AI function, in the channel of making use of the GPU’s computing power, to some extent it constitutes a purpose of the Bitcoin protocol.
Part of the Bitcoin agreement requires miners to solve complex mathematical problems. These problems are insolvable for computers on their own. Such a complex solving approach acts as a way to identify and verify transactions on the blockchain.
With such a propelling mechanism, cryptocurrencies burgeoned out, so did crypto-based transactions.
The new asset and its trading function allure investors to predict the trading trend. The most effective way to determine the future performance of a given asset is to analyse its capital flow.
By tracking the transfer of funds from known entities and diachronically comparing the transfer conditions with historical data sets, machine learning can help you predict the assets’ future value. What will be its implication?
Traders can figure out what the main players are doing in the cryptocurrency market.
For example, those “whales” may be planning a pump and dump. Such a plan would dramatically reduce the asset value. Knowing when this may occur, investors may swiftly go ashore before getting drown.
In the world of Bitcoin, the mysterious “whales” refers to those individuals with large amounts of cryptocurrencies. Bitcoin whales have been a source of speculation and anxiety for a long time. According to a research project of Chainalysis in the last year, 32 whales held more than 12,000 Bitcoins.
Or, think of another occasion, when those whales are not ready to ship now. Instead, they begin to accumulate virtual assets that have shown a recent downward trend. This downtrend allows investors to bottom out before the assets’ soaring surge.
In the field of cryptocurrency, a constellation of applications make machine learning pithy for successful transactions.
In addition, machine learning can also analyse trading behaviours. Given a group of investors, machine learning can identify the investors as samples and find out how they invest, what is investment pattern the investors follow.
Machine learning can also show its vibrancy in anti-fraud: first, the machine is so trained to grasp identified fraud patterns that it can generalise the knowledge and identify any fraudulent actions when similar patterns jump up.
Machine-learning algorithms can prevent pumps and dumps from going rage, as prevalent pumps and dumps and market manipulations may damage the entire crypto ecosystem. Machine learning, in such a way, will result in a more stable market system.
As we have seen, machine learning now booms in the field of cryptocurrency. It acts as a powerful tool to determine the on-going market trends and individuals’ and exchanges’ strategies.
This is almost the end of the speech, but the following paragraphs also deserve your special attention.
Ironically, the information about machine learning that most people hear of in the cryptocurrency field is enormously out of shape, deviated from the facts.
You will encounter all kinds of bad guys, show a “predictive model” generated as an epiphany when they were studying how machine learning works out.
At first glance, many of these models seem persuasive: they will show you a picture of a graph where they compare the results of machine predictions with the real-time market data. The comparison seemingly works out well — — It’s even an incredible match.
While any team behind these machine-learning algorithms has done a formidable work in their work, it is by any means dishonest to suggest that these algorithms should work out as a way to predict actual market transactions. In fact, their results have been revised; for example, the prediction has been delayed by a day.
In other words, these models don’t authentically reveal the correlation between the predictions and the actual results.
There may generate some models in the future, by which assets’ performance in the market may become more predictable.
Published at Tue, 24 Dec 2019 10:10:56 +0000
{flickr|100|campaign}
