was first published bloomberg.com
NORTHAMPTON, MA / ACCESSWIRE / May 12, 2023 / Bloomberg Terminal provides access to over 35 million financial instruments across all asset classes. This is a lot of data, and to harness it, AI and machine learning (ML) are playing an increasingly central role in the terminal’s continued evolution.
Machine learning is about scrutinizing data at a speed and scale far beyond what a human analyst can do. The patterns and anomalies discovered can then be used to derive powerful insights, leading to automation of all sorts of difficult or tedious tasks that previously had to be performed manually by humans.
While AI continues to fall short of human intelligence in many applications, there are areas where it significantly outperforms human agents. Machines can identify hidden trends and patterns in millions of documents, and this ability improves over time. Machines also behave in a consistent and fair manner without making the mistakes that humans inevitably make.
“Humans are great at planning things, but when it comes to making decisions, they start from scratch,” says Gideon Mann, head of ML Products & Research in the Bloomberg CTO office. “The machine does it the same way every time, so even if it makes a mistake, it makes a mistake with the same error characteristics.”
Bloomberg Terminal is currently embracing AI and ML techniques in some interesting ways, and this practice is expected to expand rapidly over the next few years. The story goes back almost 20 years…
keep humans up to date
When we started in the 80’s, data extraction was a manual process. Today, our engineers and data analysts build, train, and use AI to process unstructured data at speed and scale, so you can get to information faster.
rise of machines
Prior to the 2000s, all tasks related to data collection, analysis and distribution at Bloomberg were performed manually as the technology to automate them did not yet exist. The new millennium saw the emergence of primitive models driven by a set of human-coded if-then rules, introducing low-level automation into enterprise workflows. As this decade draws to a close, true ML is in full swing within the company. In this new approach, humans annotate data to train machines to make different associations based on labels. The machine “learns” how to make decisions based on this training data, producing more accurate results over time. This approach can be extended significantly beyond traditional rule-based programming.
Over the past decade, the use of ML applications within Bloomberg has exploded. According to James Hook, head of data at the company, there are many of his extensive AI/ML and data science applications within Bloomberg.
One is information extraction. It uses computer vision and natural language processing (NLP) algorithms to read unstructured documents (data typically arranged in forms that are difficult for machines to read) and extract semantic meaning from them. . Using these technologies, terminals can present users with insights from videos, audio, blog posts, tweets, and more.
Anju Kambadur, head of Bloomberg’s AI engineering group, explains how this works:
“Usually we start by asking questions about every document. Let’s say you have a press release. What are the entities in the document? Who is the management team involved? Who are the other companies? Do you have any supplies? Chained relationships that are evident in the documents? Then, once you have identified the entities, you can measure the salience of the relationships between the entities and target the content to specific topics. The document may be about electric vehicles, it may be about oil, it may be related to the United States, it may be related to the APAC region, all of these are “topic called a code and assigned using machine learning. ”
All this information, and much more, can be extracted from unstructured documents using natural language processing models.
Another area is quality control. Here, techniques such as anomaly detection are used to identify accuracy issues, especially in datasets. Using anomaly detection methods, the terminal can discover potential hidden investment opportunities or flag suspicious market activity. For example, if a financial analyst changes the rating of a particular stock after a company’s quarterly earnings release, anomaly detection will determine whether this is considered typical behavior, or the value this behavior presents to the analyst. It provides context as to whether there is Bloomberg customers as data points worth considering in investment decisions.
And then there’s the use of AI/ML to analyze large datasets and generate insights that reveal otherwise unobservable investment signals. One example is using highly correlated data such as credit card transactions to visualize recent company performance and consumer trends. The other analyzes and analyzes the millions of news articles that feed into the Bloomberg Terminal every day to understand the key questions and themes that drive the trading volume of a particular market or economic sector, or the securities of a particular company. to summarize.
human in loop
When we think of machine intelligence, we imagine ruthless, impartial, emotionless autonomous machines. In practice, however, ML practices are mostly human-machine teamwork. At least for now, humans are still defining ontologies and methodologies and performing annotation and quality assurance tasks. Bloomberg has moved quickly to increase the capacity of its staff to perform these tasks at scale. Machines will not replace human workers in this scenario. They are simply moving their workflow away from more tedious and repetitive tasks towards a higher level of strategic oversight.
“It’s really a transfer of human skill from manually extracting data points to thinking about defining and creating workflows,” says Mann.
Ketevan Tsereteli, Senior Scientist in Bloomberg Engineering’s Artificial Intelligence (AI) Group, explains how this transfer works in practice.
“Previously, in a manual workflow, you might have had a team of data analysts trained to find merger and acquisition news and extract relevant information from press releases. These same people now contribute to collecting and labeling this information, as well as providing feedback on the performance of ML models, and where correct assumptions are made. and points out what wrong assumptions were made. In this way, expertise in the field is gradually transferred from person to person. machine. “
Humans are needed at every step to ensure the model is performing optimally and improving over time. This includes ML engineers who build learning systems and underlying infrastructure, AI researchers and data scientists who design and implement workflows, and annotators who collect and label training data and perform quality assurance (journalists and other subject matter experts).
“Our data department has thousands of analysts with deep expertise in the areas that matter most to our clients, such as finance, law and government,” explains Tina Tseng, ML/AI Data Strategist. . “Not only do they understand the data in these areas, but they also understand how that data is being used by our customers. They work closely with our engineers and data scientists. to develop our automation solutions.”
Annotations are important not only for training a model, but also for evaluating its performance.
“We annotate the data as a set of truths, or what we call the ‘golden’ copy of the data,” says Tseng. “We can automatically compare a model’s output to its evaluation set, so we can compute statistics to quantify the model’s performance. Evaluation sets are used in both supervised and unsupervised learning.”
Check out Best Practices for Managing Data Annotation Projects, a practical guide to planning and implementing data annotation initiatives, published by Bloomberg’s CTO Office and Data Division.
read now
Visit 3blmedia.com for more multimedia and other ESG storytelling from Bloomberg.
Contact Information:
Spokesperson: Bloomberg
Website: https://www.3blmedia.com/profiles/bloomberg
Email: info@3blmedia.com
sauce: bloomberg
View the source version on accesswire.com.
https://www.accesswire.com/754570/Humans-in-the-Loop-AI-Machine-Learning-in-the-Bloomberg-terminal
