2022-006 – Multi-armed bandits with delay

State of the art decision making reinforcement-learning algorithmsTechnology is applicable to A/B testing, as well as multiple-choice problemsRobustness. Algorithms able to deal with real-life delayed outcomes, with no need for intermediate observations
  • State of the art decision making reinforcement-learning algorithms
  • Technology is applicable to A/B testing, as well as multiple-choice problems
  • Robustness. Algorithms able to deal with real-life delayed outcomes, with no need for intermediate observations

Abstract:

We have developed a method based on multi-arm bandits modeling that significantly improves decision outcomes when delayed feedback is present. Its novelty in this domain arises from its use of a predictive step which is used in evaluating decisions.

Website:
https://mcgill.flintbox.com/technologies/786ABEF493DD47A89C7D752EF69F9F3A

Advantages:

Artificial intelligence (AI) is currently revolutionizing many business practices and being used to derive new insights and drive decision-making. The retail industry is an industry which relies upon effective decision-making to generate revenue and fuel growth. Such decision-making includes determining inventory, order sizes, product pricings, and more. Currently, the in-store market segment is heavily reliant on manual decision-making. This presents a great opportunity for AI-driven decision-making for increased revenues and profits in the in-store retail industry and can be extended to online retail.

In a lab setting we have developed AI techniques based upon multi-armed bandits which optimize decision-making when there exists uncertainty in information. In particular, we have state-of-the-art technology for decision-making when decision outcomes may be delayed. Decision outcomes are often delayed in retail; there often exists a large time gap between inventory decisions and consumer purchases. We believe our technology is particularly well suited to disrupt the traditional methods of retail decision-making yielding improved profits.

Potential Applications:

RETAIL segments relying upon large volumes of transactions such as grocery stores would be particularly well-suited for adopting such technology. Retail markets concerning food products would benefit from decreased food spoilage and waste via better inventory management and pricing. Likewise, for consumers, improved pricing optimizations and inventory selection, could result in more attractive price and product offerings.

Other market applications:

  • MEDICAL/PHARMA: clinical trials planning and optimization
  • 5G/TELECOM: network resources allocation, routing
  • MANUFACTURING: cost/opportunity in sourcing tasks as well as production/manufacturing (e.g. stock/supplies management)

Contact Information:

Name : Francesco Tordini

Title :

Department :

Email: francesco.tordini@mcgill.ca

Phone: 4388811973

Address :