Label In The Loop

Label and clean your datasets for greater certainty!

Label in the Loop helps in assigning better and faster automated meaning to data (labelling), resulting in an improved machine learning model and greater consistency over time.

Technologies of the Fourth Industrial Revolution are estimated to add some $3.7t to the global economy by 2025. With more than 70% of industrial companies now experiencing tangible gains in productivity, it is simply a matter of when, and not if, the rest of industry follows suit.

One in three business leaders distrust the information available to them when making decisions, whilst IBM calculates the annual cost of poor data quality at $3.1t for the US economy alone. Most concerns regarding AI centre on data quality and the considerable cost of tagging and cleaning datasets, which together consume up to 50% of available budgets and over 80% of time.

With active learning methods driven by state of the art AI and Machine Learning techniques, Label in the Loop addresses this problem head on. LITL optimizes the process of assigning meaning to data by selecting only the relevant data for experts to tag, thus minimizing time spent on labelling. Human attention can then be redirected to more productive tasks.

Reducing man-hours spent on mundane tasks and providing greater certainty in hard data carry obvious benefits:

Cut costs
Save time
Improve quality of data
Improvements in data quality translates to better machine learning models and more accurate predictions.
Better data
Better models
Better predictions
LITL works in the following way:
  1. We store your data in our tagging system. We support many types of data - photo, video, text, structured data, sound - and/or multiple combinations of them (multimodal data).
  2. Using machine learning methods, we then select a batch of data to be tagged. The batch is selected to provide maximum added knowledge to a model trained on the tagged data.
  3. The data is then dispatched to your experts for tagging. Experts are not selected randomly - we analyse their performance, expertise, preferences and even tiredness - so as to maximise their input.
  4. We use state-of-the-art visualisation techniques in our dedicated interface to present the data in the most legible way possible.
  5. Experts then provide attributes to the data based on their expertise - we support many different types of attributes: visual, textual, relational, logical, etc.
  6. When a batch of data is tagged, various scores are calculated. A subsequent batch is selected and the process repeats.
  1. We store your data in our tagging system. We support many types of data - photo, video, text, structured data, sound - and/or multiple combinations of them (multimodal data).
  2. Using machine learning methods, we then select a batch of data to be tagged. The batch is selected to provide maximum added knowledge to a model trained on the tagged data.
  3. The data is then dispatched to your experts for tagging. Experts are not selected randomly - we analyse their performance, expertise, preferences and even tiredness - so as to maximise their input.
  4. We use state-of-the-art visualisation techniques in our dedicated interface to present the data in the most legible way possible.
  5. Experts then provide attributes to the data based on their expertise - we support many different types of attributes: visual, textual, relational, logical, etc.
  6. When a batch of data is tagged, various scores are calculated. A subsequent batch is selected and the process repeats.

Do you want to know more about BrightBox? Contact us!
Piotr Biczyk
Chief Strategy Office
Do you want to work for BrightBox? Check job offers




Project information

LITL – Labeling in the Loop. Development of an automatic and semi-automatic system for marking data in large data sets based on machine learning.

Application number: POIR.01.01.01-00-0213/19
Value of the project: 8 697 250,00 zł
Donation: 6 202 540,00 zł
Beneficiary: QED Software Sp. z o. o.
Project duration: 01.01.2020 – 30.06.2022
Project realised as a part of: Działania 1.1 Poddziałania 1.1.1 Programu Operacyjnego Inteligentny Rozwój 2014-2020 współfinansowanego ze środków Europejskiego Funduszu Rozwoju Regionalnego

Project purpose

The goal of the project is to develop and test the LITL – Labeling in the Loop system, which will ensure:

1. automation of the process of interactive marking (labeling, tagging, augmentation) of data, based on the detection of potentially relevant examples in large data sets and their ergonomic marking by field experts operating within the organization / enterprise that is the software user (client);
2. continuous self-improvement – the algorithms developed in the project will learn the specifics of user data;
3. intelligent adaptation to changing customer conditions – in the event of a change in the group of experts and thus a change in the configuration and set of competencies in the entire expert group of the client, the algorithms developed in the project will learn how to cooperate with experts in this new group;
4. minimizing the period of implementation of the project result at a new client / in the new domain – the project result will be developed and prepared for commercialization as a modular programming environment, consisting largely of universal modules, embedded within an innovative architecture integrating them with components that usually already exist in the client’s infrastructure.

The result of the project will be intended for use by those organizations and enterprises that have large and / or rapidly growing data sets, and at the same time need tools based on artificial intelligence / machine learning (AI / ML) that would allow them to fully capitalize the knowledge contained in such data.

In the project, we will develop and test AI / ML methods and algorithms implementing the abovementioned functions, and create target software with the abovementioned algorithms implemented. We will test the prototype software, preparing it for implementation in the client’s commercial activity.

🠕 Go up