AI Forecast: Collaborative Machine Learning without Centralized Trading Data


The past few years have seen rapid advances in machine learning, with new technologies achieving dramatic improvements in technical performance. But we can go beyond optimizing objective functions. By building AI systems with users in mind from the ground up, we open up entire new areas of finance.

AI Forecast is devoted to advancing the research and design of people-centric AI systems. We're interested in the full spectrum of human trading interaction with machine intelligence, from supporting engineers to understanding everyday trading experiences with AI. 

A new paradigm for learning

" Man is a gaming animal. He must always be trying to get the better in something or other."

Charles Lamb

In our experiment, we focus on an approach known as Decision making using game theory. We apply principles from game theory to model the relationships between rating actions, news, market signals and decision making. We postulate the use of design capability indices to facilitate the teams making a ranged set of decisions, instead of specific ones. 

In the context of stock price realization, a game is a decision making process between multiple investors each of which controls a subset of design variables and seeks to minimize its cost function subject to future forecast constraints. That is, investors act like players in a game; they cooperate to achieve a set of overall goals. 

Modern machine learning models are highly flexible but lack transparency. Can we devise methods to explain the predictions of such models, without restricting their expressiveness? Can we do so even if we don't know anything about their architecture, i.e., if they are "black-boxes"? In this project, we are developing methods for explaining the predictions made rather than constraining the models themselves to be interpretable. We are particularly interested in providing explanations for the predictions of complex machine learning models that operate on structured data, such as sentences, trees or graphs. For example, we use statistical input-output analysis to learn to interpret predictions of sequence-to-sequence models, such as those used in machine translation and dialogue systems. 

Our goal is to do fundamental research, invent new technology, and create frameworks for forecast in order to drive a human-centered approach to artificial intelligence. 

References

Deep Reinforcement Learning in Large Discrete Action Spaces
Applying reasoning in an environment with a large number of discrete actions to bring reinforcement learning to a wider class of problems.
View paper

Deep Reinforcement Learning with Attention for Slate Markov Decision Processes with High-Dimensional States and Actions
Introducing slate Markov Decision Processes (MDPs), a formulation that allows reinforcement learning to be applied to recommender system problems.

View paper

Massively Parallel Methods for Deep Reinforcement Learning
Presenting the first massively distributed architecture for deep reinforcement learning.
View paper

Adaptive Lambda Least-Squares Temporal Difference Learning
Learning to select the best value of λ (which controls the timescale of updates) for TD(λ) to ensure the best result when trading off bias against variance. 
View paper

Learning from Demonstrations for Real World Reinforcement Learning
Presenting Deep Q-learning from Demonstrations (DQfD), an algorithm that leverages data from previous control of a system to accelerate learning.
View paper

Value-Decomposition Networks For Cooperative Multi-Agent Learning
Studying the problem of cooperative multi-agent reinforcement learning with a single joint reward signal.
View paper

Many Paths to Equilibrium: GANs Do Not Need to Decrease a Divergence At Every Step
Demonstrating an alternative view of the training of GANs.
View paper

Risk-Constrained Reinforcement Learning with Percentile Risk Criteria
Presenting efficient reinforcement learning algorithms for risk-constrained Markov decision processes (MDPs) and demonstrating their effectiveness in an optimal stopping problem and an online marketing application.
View paper



We work across the world

From London to San Francisco, to our home base in (Saint Helier) Jersey, we’re looking for extraordinary and creative scientists to help us drive the field forward.

AC Investment Inc. currently does not act as an equities executing broker or route orders containing equities securities. If AC Invest’s business model were to change and it begins routing non-directed orders in NMS securities, it will comply with the disclosure requirement of Rule 606.

77 Massachusetts Avenue Cambridge, MA 02139 617-253-1000 pr@ademcetinkaya.com