-
Sparse Explanations of Neural Networks Using Pruned Layer-Wise Relevance Propagation
Authors:
Paulo Yanez Sarmiento,
Simon Witzke,
Nadja Klein,
Bernhard Y. Renard
Abstract:
Explainability is a key component in many applications involving deep neural networks (DNNs). However, current explanation methods for DNNs commonly leave it to the human observer to distinguish relevant explanations from spurious noise. This is not feasible anymore when going from easily human-accessible data such as images to more complex data such as genome sequences. To facilitate the accessib…
▽ More
Explainability is a key component in many applications involving deep neural networks (DNNs). However, current explanation methods for DNNs commonly leave it to the human observer to distinguish relevant explanations from spurious noise. This is not feasible anymore when going from easily human-accessible data such as images to more complex data such as genome sequences. To facilitate the accessibility of DNN outputs from such complex data and to increase explainability, we present a modification of the widely used explanation method layer-wise relevance propagation. Our approach enforces sparsity directly by pruning the relevance propagation for the different layers. Thereby, we achieve sparser relevance attributions for the input features as well as for the intermediate layers. As the relevance propagation is input-specific, we aim to prune the relevance propagation rather than the underlying model architecture. This allows to prune different neurons for different inputs and hence, might be more appropriate to the local nature of explanation methods. To demonstrate the efficacy of our method, we evaluate it on two types of data, images and genomic sequences. We show that our modification indeed leads to noise reduction and concentrates relevance on the most important features compared to the baseline.
△ Less
Submitted 22 April, 2024;
originally announced April 2024.
-
On the area swept by a biased diffusion till its first-exit time: Martingale approach and gambling opportunities
Authors:
Yonathan Sarmiento,
Debraj Das,
Édgar Roldán
Abstract:
Using martingale theory, we compute, in very few lines, exact analytical expressions for various first-exit-time statistics associated with one-dimensional biased diffusion. Examples include the distribution for the first-exit time from an interval, moments for the first-exit site, and functionals of the position, which involve memory and time integration. As a key example, we compute analytically…
▽ More
Using martingale theory, we compute, in very few lines, exact analytical expressions for various first-exit-time statistics associated with one-dimensional biased diffusion. Examples include the distribution for the first-exit time from an interval, moments for the first-exit site, and functionals of the position, which involve memory and time integration. As a key example, we compute analytically the mean area swept by a biased diffusion until it escapes an interval that may be asymmetric and have arbitrary length. The mean area allows us to derive the hitherto unexplored cross-correlation function between the first-exit time and the first-exit site, which vanishes only for exit problems from symmetric intervals. As a colophon, we explore connections of our results with gambling, showing that betting on the time-integrated value of a losing game it is possible to design a strategy that leads to a net average win.
△ Less
Submitted 10 May, 2024; v1 submitted 31 December, 2023;
originally announced January 2024.
-
Human perceptual decision making of nonequilibrium fluctuations
Authors:
Aybüke Durmaz,
Yonathan Sarmiento,
Gianfranco Fortunato,
Debraj Das,
Mathew Ernst Diamond,
Domenica Bueti,
Édgar Roldán
Abstract:
Perceptual decision-making frequently requires making rapid, reliable choices upon encountering noisy sensory inputs. To better define the statistical processes underlying perceptual decision-making, here we characterize the choices of human participants visualizing a system of nonequilibrium stationary physical dynamics and compare such choices to the performance of an optimal agent computing Wal…
▽ More
Perceptual decision-making frequently requires making rapid, reliable choices upon encountering noisy sensory inputs. To better define the statistical processes underlying perceptual decision-making, here we characterize the choices of human participants visualizing a system of nonequilibrium stationary physical dynamics and compare such choices to the performance of an optimal agent computing Wald's sequential probability ratio test (SPRT). Participants viewed movies of a particle endowed with drifted Brownian dynamics and had to judge the motion as leftward or rightward. Overall, the results uncovered fundamental performance limits, consistent with recently established thermodynamic trade-offs involving speed, accuracy, and dissipation. Specifically, decision times are sensitive to entropy production rates. Moreover, to achieve a given level of observed accuracy, participants require more time than predicted by SPRT, indicating suboptimal integration of available information. In view of such suboptimality, we develop an alternative account based on evidence integration with a memory time constant. Setting the time constant proportionately to the deviation from equilibrium in the stimuli significantly improved trial-by-trial predictions of decision metrics with respect to SPRT. This study shows that perceptual psychophysics using stimuli rooted in nonequilibrium physical processes provides a robust platform for understanding how the brain takes decisions on stochastic information inputs.
△ Less
Submitted 22 November, 2023; v1 submitted 21 November, 2023;
originally announced November 2023.
-
Detection of arbitrage opportunities in multi-asset derivatives markets
Authors:
Antonis Papapantoleon,
Paulo Yanez Sarmiento
Abstract:
We are interested in the existence of equivalent martingale measures and the detection of arbitrage opportunities in markets where several multi-asset derivatives are traded simultaneously. More specifically, we consider a financial market with multiple traded assets whose marginal risk-neutral distributions are known, and assume that several derivatives written on these assets are traded simultan…
▽ More
We are interested in the existence of equivalent martingale measures and the detection of arbitrage opportunities in markets where several multi-asset derivatives are traded simultaneously. More specifically, we consider a financial market with multiple traded assets whose marginal risk-neutral distributions are known, and assume that several derivatives written on these assets are traded simultaneously. In this setting, there is a bijection between the existence of an equivalent martingale measure and the existence of a copula that couples these marginals. Using this bijection and recent results on improved Fréchet-Hoeffding bounds in the presence of additional information on functionals of a copula by Lux and Papapantoleon [18], we can extend the results of Tavin [33] on the detection of arbitrage opportunities to the general multi-dimensional case. More specifically, we derive sufficient conditions for the absence of arbitrage and formulate an optimization problem for the detection of a possible arbitrage opportunity. This problem can be solved efficiently using numerical optimization routines. The most interesting practical outcome is the following: we can construct a financial market where each multi-asset derivative is traded within its own no-arbitrage interval, and yet when considered together an arbitrage opportunity may arise.
△ Less
Submitted 20 November, 2021; v1 submitted 14 February, 2020;
originally announced February 2020.
