Research-driven start-up companies

The authors argue that Silicon Valley is returning to the tradition of research and development, and that startups should give up the pursuit of a perfect start, start by releasing the available products as soon as possible, and after the user and venture capital, further research and development is not too late.

The foundation of Silicon Valley is research, and the WEB boom has transformed it into consumer-media software. Companies such as http://www.aliyun.com/zixun/aggregation/1560.html ">facebook and Twitter are starting with simple programs, getting investments and recruiting researchers."

However, I think we may be in the midst of a period of enlightenment in data and research revival.

We are seeing a growing number of research-driven data-type startups. We're seeing a growing number of startups like Facebook and Twitter that don't start with research and data, but then need to research or data to support personalization, precision advertising delivery, product referrals, premium products or other forms of intelligence to make revenue.

If the assumption is correct, then you need to know how to do this kind of startup.

Through the product to solve the problem, based on research, through data processing to promote

The existence value of this kind of start-up company is at a glance.

Collect and process data, then extract information from it, and study to obtain useful information to develop products that can solve a problem.

Problem <-Product <-Intelligence <-Research <-Information <-processing <-data

At some point, we only think of the "information" phase, there is no relationship, the acquisition of "intelligence" still needs in-depth work.

Skills

We need to rely on three major skills.

researchers: machine learning, statistics, mathematics, computer science. System hackers: Computer scientists and engineers focused on data storage, transmission, queuing, and processing, in many cases, require the skills of a distributed system. Front-End engineers: Designers, interactive designers, JavaScript Masters, user experience.

Researchers and front-end engineers need to focus on the product, while the three have to focus on the data.

Through the product decomposition of complexity

There is a data moat that is far from enough.

If you are business-to-business, you can use the API to throw the data to the user. If you're a consumer, consumers don't want any data, they just want the problem solved.

Even if most business-to-business, the user needs is not raw data, but processed data. They want a signal of when to act, not an obscure probability distribution.

People usually do not turn to probability theory. See Kahneman and Tversky's studies, as well as their expected theories.

For the Business-to-consumer, you have to have a lovely product. Then you need to get information from the data and confidently expect yourself to be able to process the information into useful information.

So the big winner in this field is the ability to load a huge amount of data and complexity into a minimalist interface. See figure below:p

Find the "good enough" model first.

Research-driven data start-ups tend to be heavily constrained in their resources. So, first develop a "good enough" and simple model, using this model to identify problems, and then get the attention of customers and investors financing, as well as such things.

For such startups, it should be done and come up with the product before the best solution is released. With a good enough plan, we can continue to study and gradually become the best solution.

This is an important concept and it is advisable to consider it as early as possible. If, at the same risk, you can beat the S & P 500 index with only small marginal gains, you will be a hero. For other problems, as long as you work hard enough, nature can solve.

If the results must be accurate, then perhaps the problem you are trying to solve is not suitable for research-driven startups. If your project only needs a little improvement to be productive, it's a good place to start startups.

Day after day, you always do better-the perfect start isn't enough to put you on the whole bet.

Starting from a single data source

If you integrate many different data sources into a single view to create your eigenvector, then you might want to use a single data source as the basis for the model, and then load the other data sources again.

The pattern of many problems is interlinked: a major dense data source and several sparse data sources that add to the master source.

If you consolidate too much data at once, you may find yourself overwhelmed by the complexity of data processing and conversion, which can also hurt your research skills. It may also limit your ability to extract information from a data source, because constantly modifying the data makes it difficult to focus on extracting information from each data source.

Lessons learned from software development

Early release, frequent release, multiple evaluation. Yes-you can continue to study while doing this.

The idea that you can arbitrarily set future research goals is sheer nonsense. Step-this method is better.

The pace may vary, the time may be different, and there may be many difficulties, but you can still step in.

Is the research that takes me into agile and TDD (editor: Test-driven Development, Test-driven Development). I have been doing research since 2003. TDD is a science-come up with your hypothesis, find a way to test it, and then test it.

Select your metrics and test methods. What is good enough? When can we reach the point of diminishing returns?

Don't think about the things you can't finish after you die. You have to keep telling yourself.

Hypothesis Testing

Keep in mind that in startups, everything is hypothetical, and your job is to test the assumptions.

How much data can you extract from a data source? Maybe you have a lot of data, but too much noise and too little value. How sparse is the data? That very informative but rare source of data is also unhelpful. Can you find a model that is simple enough to start quickly and also attract users? Can you take advantage of the income from the early "good enough" model to expand your business or study diligently? Can your model be product or fit into a service that people really care about?

Create Solutions to problems

Think back to the research-driven data startups that get the data, get the information by processing the data, then work out useful information on it, then make it into a product and solve a problem.

Also, remember that in some cases we can only get information, but that's enough. We still have a lot of problems. Sometimes, the intelligence part is not very necessary, rather early to make a product, may wish to move the solution to the future.

To solve the problem for people, please do some research.

[Original link; Author: Bradford Cross]

Source: https://apple4.us/2010/07/research-driven-startups.html