风险业务: 利用 AI 来评估和限制风险
这篇文章来自 nvidia.com。原始 url 是: https://blogs.nvidia.com/blog/2018/04/19/ai-assess-merger-risk/
以下内容由机器翻译生成。如果您觉得可读性不好, 请阅读原文或 点击这里.
Assessing risk is a concern in most industries, although perhaps never more so than following a merger or acquisition. It turns out AI may be just the tool to help.
GPU 技术会议 attendees last month got a high-level education in how AI can bring speed and precision to this process from Congruity360, a Massachusetts-based data management consultancy.
Mitigating Risk in M&A
Two people deciding to live together can see all of their belongings and make quick decisions about what to do with them. But a company that’s joined another is almost always saddled with data that’s not so easy to categorize.
While AI may not be well suited to analyzing text-based documents, Congruity360 has developed a method for parsing text data with GPU-powered machine learning.
“GPUs are not going to operate on text,” said Chris Ryan, vice president of sales engineering at Congruity360. “We need to come up with a mathematical representation of text documents.”
Doing so has allowed Congruity360 to classify unstructured documents based on whether they look the same or contain some of the same keywords. At its essence, the company’s work involves taking data it knows nothing about — “dark data,” as Ryan called it — and assigning high-level headers so it can separate the data into buckets related to topics such as invoices, taxes, intellectual property or even code.
The result is a visual representation that groups data in topical clusters, some of which stand on their own and some of which overlap. Companies can use this method to zero in on clusters of riskier documents, such as those that have regulatory implications.
Turning Data into Useful Information
Congruity360’s approach starts with the assumption that as much as 80 percent of all corporate data is unstructured, and seeks to answer the question, how can GPUs help machine learning turn raw text into information?
Most obviously, GPUs bring speed to the equation.
“If you’re a data scientist and you want to do this, you don’t want to wait weeks and weeks for models to run,” said Jonathan Bailey, vice president of analytics at Congruity360.
Speeding up the process translates to identifying — and mitigating — risks sooner. M&A activity involves working with legal teams, which are typically most concerned with ensuring that data is defensible. Congruity360 uses GPUs to perform comparisons of documents and compute their defensibility. It’s a process took four weeks using CPUs, and now unfolds in just 20 seconds on GPUs.
“We’re just trying to give users a tool to learn about data,” said Bailey. “We want to see if there’s any risky data in there.”