New method tackles catastrophic forgetting and order-sensitivity in continual learning.
problem Catastrophic forgetting and order-sensitivity in continual learning.
method Additive Parameter Decomposition (APD) to represent task parameters as a sum of shared and adaptive parts.
result Significantly outperforms state-of-the-art methods in accuracy, scalability, and order-robustness.
Researchers quantify risk exposure and sensitivities in financial markets under model uncertainty.
problem Optimizing investment and pricing under model uncertainty in financial markets.
method Distributionally robust optimization, Wasserstein ball, first-order sensitivity analysis.
result Sensitivities of value function, investment policy, and marginal prices to model uncertainty can be non-monotonic.
Transformers for binary decisions are sensitive to evidence order, leading to unreliable outcomes.
problem Order sensitivity in Transformers for binary decisions leads to unreliable outcomes.
method Formalized an expectation-realization gap and developed QMV and EDFL bounds.
result Uniform permutation mixtures reduce dispersion and improve reliability.
New metric derived for robust optimization in stochastic control problems.
problem Non-parametric uncertainty in multiperiod stochastic control problems.
method Derived a new metric, adapted (p,∞)--Wasserstein distance, and used dynamic programming principle. result Dynamic programming principle for DRO problems with semi-separable cost functions.
Banks must manage their trading books, not just value them. Pricing includes valuation adjustments collectively known as XVA (at least credit, funding, capital and tax), so management must also include XVA. In trading book management we focus on pricing, hedging, and allocation of prices or hedging costs to desks on an…
Deep BSDE method for pricing and hedging complex financial portfolios.
problem Simultaneous pricing and delta-gamma hedging of large portfolios of multi-asset Bermudan options.
method Discretely reflected BSDEs, One Step Malliavin scheme, neural network regression Monte Carlo method.
result Efficient and accurate pricing and hedging strategies for high-dimensional portfolios.
Introduces TPV to analyze model robustness without labels.
problem Analyzing post-training robustness of machine learning models.
method Parameter perturbations and test prediction variance (TPV) as a unifying framework.
result TPV connects various perturbations under a single lens, providing insights into model stability.
Proposes a new framework for invariant quadratic P&L predictions in option books.
problem Inconsistent second-order P&L predictions across different factor parameterizations.
method Local, model-agnostic framework using covariant Hessian defined by an affine connection.
result Coordinate-invariant quadratic P&L predictions that match desk targets.
PARD generates graphs efficiently and invariantly to node ordering.
problem Graph generation sensitivity to node ordering.
method Integrates autoregressive and diffusion models with a partial order for nodes and edges.
result PARD achieves state-of-the-art performance on molecular and non-molecular datasets.
We formalize causal separation in portfolio theory, deriving a closed-form projected Markowitz solution.
problem Portfolio optimization under causal separation conditions.
method Derive a closed-form solution for portfolio optimization using causal separation conditions.
result A closed-form projected Markowitz solution is derived under causal separation conditions.
Optimizes neural networks for solving problems with pruning and ensembles of minimal structures.
problem Improving neural network performance and interpretability.
method Pruning neural networks based on the principle of controlling training and pruning, using sensitivity indicators and logically transparent NN.
result Ensemble of minimal neural networks provides diverse forecasting algorithms and identifies areas for further data collection.
Transformers approximate Bayesian posteriors but not exactly.
problem Bayesian accounts of in-context learning face challenges due to task-preserving order changes in transformers.
method Showed that excess prequential code length is exactly cumulative predictive KL, decomposing expected regret into order-averaged predictor and order-averaging gain.
result Transformers approximate Bayesian posteriors but not exactly, priced by log loss.