AIFI - Artificial Intelligence Finance Institute

Harry Mendell and I wrote a paper: "Large Language Models in Financial Services: FAIR — A Framework for Implementation, Risk Mitigation, and Remediation?" You can download it here : https://lnkd.in/du-XA_iP AIFI - Artificial Intelligence Finance Institute Global AI Captide

I wrote a paper: "Large Language Model Training and Reinforcement Learning" https://lnkd.in/dYA_vapV AIFI - Artificial Intelligence Finance Institute Global AI Captide

Fran?ois Charton gave a talk in our AIFI - Artificial Intelligence Finance Institute Bootcamp 2025 about these great papers: PatternBoost: Constructions in Mathematics with a Little Help from AI Authors:?Fran?ois Charton,?Jordan S. Ellenberg,?Adam Zsolt Wagner,?Geordie Williamson - 2024 Abstract:?We introduce PatternBoost, a flexible method for finding interesting constructions in mathematics. Our algorithm alternates between two phases. In the first ``local'' phase, a classical search algorithm is used to produce many desirable constructions. In the second ``global'' phase, a transformer neural network is trained on the best such constructions. Samples from the trained transformer are the… https://lnkd.in/diQ2qrBH Global Lyapunov functions: a long-standing open problem in mathematics, with symbolic transformers - 2024 Authors:?Alberto Alfarano,?Fran?ois Charton,?Amaury Hayat Abstract:?Despite their spectacular progress, language models still struggle on complex reasoning tasks, such as advanced mathematics. We consider a long-standing open problem in mathematics: discovering a Lyapunov function that ensures the global stability of a dynamical system. This problem has no known general solution, and algorithmic solvers only exist for some small polynomial systems. We propose a ne… https://lnkd.in/dtTBagJT Emergent properties with repeated examples Authors:?Fran?ois Charton,?Julia Kempe - 2024 Abstract:?We study the performance of transformers as a function of the number of repetitions of training examples with algorithmically generated datasets. On three problems of mathematics: the greatest common divisor, modular multiplication, and matrix eigenvalues, we show that for a fixed number of training steps, models trained on smaller sets of repeated examples outperform models trained on larger sets… https://lnkd.in/d6YAsuzz #aiformathematics

Trust Region Policy Optimization John Schulman,?Sergey Levine,?Philipp Moritz,?Michael I. Jordan,?Pieter Abbeel - 2015 We describe an iterative procedure for optimizing policies, with guaranteed monotonic improvement. By making several approximations to the theoretically-justified procedure, we develop a practical algorithm, called Trust Region Policy Optimization (TRPO). This algorithm is similar to natural policy gradient methods and is effective for optimizing large nonlinear policies such as neural networks. Our experiments demonstrate its robust performance on a wide variety of tasks: learning simulated robotic swimming, hopping, and walking gaits; and playing Atari games using images of the screen as input. Despite its approximations that deviate from the theory, TRPO tends to give monotonic improvement, with little tuning of hyperparameters. AIFI - Artificial Intelligence Finance Institute Global AI Captide https://lnkd.in/dsaT4bY8

Continuous control with deep reinforcement learning Timothy P. Lillicrap,?Jonathan J. Hunt,?Alexander Pritzel,?Nicolas Heess,?Tom Erez,?Yuval Tassa,?David Silver,?Daan Wierstra - 2015 We adapt the ideas underlying the success of Deep Q-Learning to the continuous action domain. We present an actor-critic, model-free algorithm based on the deterministic policy gradient that can operate over continuous action spaces. Using the same learning algorithm, network architecture and hyper-parameters, our algorithm robustly solves more than 20 simulated physics tasks, including classic problems such as cartpole swing-up, dexterous manipulation, legged locomotion and car driving. Our algorithm is able to find policies whose performance is competitive with those found by a planning algorithm with full access to the dynamics of the domain and its derivatives. We further demonstrate that for many of the tasks the algorithm can learn policies end-to-end: directly from raw pixel inputs. AIFI - Artificial Intelligence Finance Institute Global AI Captide https://lnkd.in/dXcGbceQ

Simple Statistical Gradient-Following Algorithms for Connectionist Reinforcement Learning Ronald J. Williams - 1992 Northeastern University This article presents a general class of associative reinforcement learning algorithms for connectionist networks containing stochastic units. These algorithms, called REINFORCE algorithms, are shown to make weight adjustments in a direction that lies along the gradient of expected reinforcement in both immediate-reinforcement tasks and certain limited forms of delayed-reinforcement tasks, and they do this without explicitly computing gradient estimates or even storing information from which such estimates could be computed. Specific examples of such algorithms are presented, some of which can be naturally integrated with existing algorithms while others may be better left functioning in their own right. Also given are results that show how such algorithms can bear a close relationship to certain backpropagation methods when a limited discounting of the number of expected reinforcements is used. We conclude with a brief discussion about how these algorithms might be used to help develop shallow but potentially more powerful reinforcement learning algorithms. AIFI - Artificial Intelligence Finance Institute Global AI Captide

A Theory of the Learnable Leslie G. Valiant - 1984 Abstract Humans appear to be able to learn new concepts without needing to be programmed explicitly in any conventional sense. In this paper we regard learning as the phenomenon of knowledge acquisition in the absence of explicit programming. We give a precise methodology for studying this phenomenon from a computational viewpoint. It consists of choosing an appropriate information gathering mechanism, the learning protocol, and exploring the class of concepts that can be learned using it in a reasonable (polynomial) number of steps. Although inherent algorithmic complexity appears to set serious limits to the range of concepts that can be learned, we show that there are some important nontrivial classes of propositional concepts that can be learned in a realistic sense. AIFI - Artificial Intelligence Finance Institute Global AI Captide

Today, I presented my work on Common Causal Sensitivity-Based Portfolio Optimization at the AIFI Winter Bootcamp 2025. The presentation included new content expanding on the original 2022 paper, addressing questions collected from past conferences and talks. I introduced two methods for optimal portfolio driver selection based on the Principle of Commonality. One approach uses correlation as a proxy for driver commonality—effectively regularizing the optimization by favoring drivers that are more broadly shared across assets. The second method is based on Bayesian networks, where, under linear modeling assumptions, the score to minimize becomes the covariance—aligning with the notion that in normally distributed systems, independence implies zero covariance. I also provided a broader conceptual overview, placing this framework within the historical evolution of portfolio theory from Markowitz to the present, highlighting practical market constraints that limit the feasibility of some existing theoretical models. Finally, I presented a simplified, visual explanation of the Principle of Commonality—grounded in probabilistic common causality, drawing on foundational ideas from Reichenbach (1956) and MPT (1952). It is shown that for a sensitivity space to exist and allow for maximum idiosyncratic diversification while incorporating systematic risk, the selected drivers must be both common and causal (Slides 26–29). All of this is included in the original publication, which comes with open-source code and data. https://lnkd.in/d3szFear link to the original paper: https://lnkd.in/dzTWXudv

The 8th Women in Quantitative Finance (WQF) Conference is scheduled for Thursday, April 24, 2025, at Level 39, One Canada Square, Canary Wharf, London. This annual event has become a cornerstone for professionals in the field, offering a unique platform to engage with cutting-edge research, network with industry leaders, and participate in dynamic discussions. https://lnkd.in/gxtuZEv Why Attend the 8th WQF Conference? Distinguished Speaker Line-Up: The conference boasts a roster of esteemed speakers from leading financial institutions. These experts will share insights on topics ranging from, Times Series, Interest rates and inflation, Implied volatility and Advancements in product valuation methodologies. Cutting-Edge Research: Attendees will have the opportunity to delve into the latest quantitative finance research, gaining exposure to innovative methodologies and applications that are shaping the future of the industry. Networking Opportunities: The conference provides a platform to connect with professionals across various sectors of quantitative finance, fostering relationships that can lead to collaborations, mentorship, and career advancement. Engaging Panel Discussions: Participate in passionate panel discussions addressing current challenges and opportunities within the field, offering diverse perspectives and solutions. Commitment to Diversity and Inclusion: The WQF Conference is dedicated to promoting the advancement of women in quantitative finance, creating an inclusive environment that encourages participation and growth. Registration Details Given the high demand and the fact that the previous conference sold out, early registration is recommended. A nominal attendance fee of ￡30 (￡15 for full-time academics) has been introduced to manage attendance effectively. Additionally, there’s a special offer where, if two colleagues attend, the third can join for free. Don’t miss this opportunity to be part of a significant event that contributes to the advancement of women in quantitative finance. Register promptly to secure your place at the 8th WQF Conference. Speakers: Wafaa Schiefler, Mirela Predescu, Milena Vuletic, Antonia Lim, Milena Imamovic-Tomasovic, Svetlana Borovkova, Amira AKKARI, Sanja Hukovic, Laura Ballotta, Aakriti Mittal, Maria M., Irene Perdomo, Imane B. Sponsors: JPMorganChase, MatLogica.

Asynchronous Methods for Deep Reinforcement Learning - 2016 Volodymyr Mnih,?Adrià Puigdomènech Badia,?Mehdi Mirza,?Alex Graves,?Timothy P. Lillicrap,?Tim Harley,?David Silver,?Koray Kavukcuoglu We propose a conceptually simple and lightweight framework for deep reinforcement learning that uses asynchronous gradient descent for optimization of deep neural network controllers. We present asynchronous variants of four standard reinforcement learning algorithms and show that parallel actor-learners have a stabilizing effect on training allowing all four methods to successfully train neural network controllers. The best performing method, an asynchronous variant of actor-critic, surpasses the current state-of-the-art on the Atari domain while training for half the time on a single multi-core CPU instead of a GPU. Furthermore, we show that asynchronous actor-critic succeeds on a wide variety of continuous motor control problems as well as on a new task of navigating random 3D mazes using a visual input. https://lnkd.in/dXKT3qG4 AIFI - Artificial Intelligence Finance Institute Global AI Captide