scholarly journals Gradient-Based Optimization for Bayesian Preference Elicitation

2020 ◽  
Vol 34 (06) ◽  
pp. 10292-10301
Author(s):  
Ivan Vendrov ◽  
Tyler Lu ◽  
Qingqing Huang ◽  
Craig Boutilier
Keyword(s):  
Gradient Methods ◽  
User Preferences ◽  
Large Item ◽  
Bayesian Criterion ◽  
Gradient Based ◽  
Expected Value Of Information ◽  
Item Attributes ◽  
Explicit Enumeration ◽  
Continuous Formulation ◽  
Modern Machine

Effective techniques for eliciting user preferences have taken on added importance as recommender systems (RSs) become increasingly interactive and conversational. A common and conceptually appealing Bayesian criterion for selecting queries is expected value of information (EVOI). Unfortunately, it is computationally prohibitive to construct queries with maximum EVOI in RSs with large item spaces. We tackle this issue by introducing a continuous formulation of EVOI as a differentiable network that can be optimized using gradient methods available in modern machine learning computational frameworks (e.g., TensorFlow, PyTorch). We exploit this to develop a novel Monte Carlo method for EVOI optimization, which is much more scalable for large item spaces than methods requiring explicit enumeration of items. While we emphasize the use of this approach for pairwise (or k-wise) comparisons of items, we also demonstrate how our method can be adapted to queries involving subsets of item attributes or “partial items,” which are often more cognitively manageable for users. Experiments show that our gradient-based EVOI technique achieves state-of-the-art performance across several domains while scaling to large item spaces.

scholarly journals Evolutionary Social Poisson Factorizationfor Temporal Recommendation

2021 ◽  
Vol 14 (1) ◽  
Author(s):  
ChunYan Yin ◽  
YongHeng Chen ◽  
Wanli Zuo
Keyword(s):  
Recommendation Systems ◽  
Predictive Performance ◽  
User Preferences ◽  
Social Model ◽  
User Preference ◽  
Latent Factors ◽  
Factorization Model ◽  
Real World Datasets ◽  
Item Attributes ◽  
The Impact

AbstractPreference-based recommendation systems analyze user-item interactions to reveal latent factors that explain our latent preferences for items and form personalized recommendations based on the behavior of others with similar tastes. Most of the works in the recommendation systems literature have been developed under the assumption that user preference is a static pattern, although user preferences and item attributes may be changed through time. To achieve this goal, we develop an Evolutionary Social Poisson Factorization (EPF$$\_$$ _ Social) model, a new Bayesian factorization model that can effectively model the smoothly drifting latent factors using Conjugate Gamma–Markov chains. Otherwise, EPF$$\_$$ _ Social can obtain the impact of friends on social network for user’ latent preferences. We studied our models with two large real-world datasets, and demonstrated that our model gives better predictive performance than state-of-the-art static factorization models.

scholarly journals A Penalized Linear and Nonlinear Combined Conjugate Gradient Method for the Reconstruction of Fluorescence Molecular Tomography

2007 ◽  
Vol 2007 ◽  
pp. 1-19 ◽  
Author(s):  
Shang Shang ◽  
Jing Bai ◽  
Xiaolei Song ◽  
Hongkai Wang ◽  
Jaclyn Lau
Keyword(s):  
Conjugate Gradient Method ◽  
Conjugate Gradient ◽  
Gradient Method ◽  
Optimization Problems ◽  
Gradient Methods ◽  
Reconstruction Algorithms ◽  
Fluorescence Molecular Tomography ◽  
Nonlinear Conjugate Gradient Method ◽  
Gradient Based ◽  
Linear And Nonlinear

Conjugate gradient method is verified to be efficient for nonlinear optimization problems of large-dimension data. In this paper, a penalized linear and nonlinear combined conjugate gradient method for the reconstruction of fluorescence molecular tomography (FMT) is presented. The algorithm combines the linear conjugate gradient method and the nonlinear conjugate gradient method together based on a restart strategy, in order to take advantage of the two kinds of conjugate gradient methods and compensate for the disadvantages. A quadratic penalty method is adopted to gain a nonnegative constraint and reduce the illposedness of the problem. Simulation studies show that the presented algorithm is accurate, stable, and fast. It has a better performance than the conventional conjugate gradient-based reconstruction algorithms. It offers an effective approach to reconstruct fluorochrome information for FMT.

Pulse Modulation - A Novel Approach to Gradient-Based Flow Injection Techniques

2001 ◽  
Vol 66 (8) ◽  
pp. 1219-1237 ◽  
Author(s):  
Armando Herbelin ◽  
Jaromir Ruzicka
Keyword(s):  
Flow Injection ◽  
Small Volume ◽  
Gradient Methods ◽  
Flow Conditions ◽  
Pulse Modulation ◽  
Novel Approach ◽  
Gradient Based ◽  
Injection Techniques ◽  
Computer Controlled ◽  
Single Standard

Development of a novel system for generation of gradients in flow injection analysis by pulse modulation is described. These user-selectable gradients are created by computer-controlled mixing of two solutions with a total volume as low as 75 μl and can be delivered under incremental or continuous flow conditions. Applications such as automated, single-standard instrument calibration are expected to benefit from high-precision linear gradients (r2 = 0.99989, n = 55). Gradient methods in biochemisty and immunology such as kinetic measurement of biomolecular interactions will benefit from the small volume of these gradients, especially for analytes with limited availability.

Robust inversion of seismic data using the Huber norm

Geophysics ◽  
2003 ◽  
Vol 68 (4) ◽  
pp. 1310-1319 ◽  
Author(s):  
Antoine Guitton ◽  
William W. Symes
Keyword(s):  
Seismic Data ◽  
Field Data ◽  
Gradient Methods ◽  
Nonlinear Problems ◽  
Velocity Analysis ◽  
Error Measures ◽  
Robust Model ◽  
Least Squares Fit ◽  
Gradient Based ◽  
Quasi Newton

The “Huber function” (or “Huber norm” ) is one of several robust error measures which interpolates between smooth (ℓ2) treatment of small residuals and robust (ℓ1) treatment of large residuals. Since the Huber function is differentiable, it may be minimized reliably with a standard gradient‐based optimizer. We propose to minimize the Huber function with a quasi‐Newton method that has the potential of being faster and more robust than conjugate‐gradient methods when solving nonlinear problems. Tests with a linear inverse problem for velocity analysis with both synthetic and field data suggest that the Huber function gives far more robust model estimates than does a least‐squares fit with or without damping.

scholarly journals Diversity-Inducing Policy Gradient: Using Maximum Mean Discrepancy to Find a Set of Diverse Policies

2019 ◽  
Author(s):  
Muhammad Masood ◽  
Finale Doshi-Velez
Keyword(s):  
Reinforcement Learning ◽  
Optimization Technique ◽  
Gradient Methods ◽  
Domain Expert ◽  
Learning Methods ◽  
Maximum Mean Discrepancy ◽  
Optimal Policies ◽  
Policy Gradient ◽  
Gradient Based ◽  
The Difference

Standard reinforcement learning methods aim to master one way of solving a task whereas there may exist multiple near-optimal policies. Being able to identify this collection of near-optimal policies can allow a domain expert to efficiently explore the space of reasonable solutions.  Unfortunately, existing approaches that quantify uncertainty over policies are not ultimately relevant to finding policies with qualitatively distinct behaviors.  In this work, we formalize the difference between policies as a difference between the distribution of trajectories induced by each policy, which encourages diversity with respect to both state visitation and action choices.  We derive a gradient-based optimization technique that can be combined with existing policy gradient methods to now identify diverse collections of well-performing policies.  We demonstrate our approach on benchmarks and a healthcare task.

Recent Advances in Gradient Based Unconstrained Optimization Techniques for Large Problems

1983 ◽  
Vol 105 (2) ◽  
pp. 155-159 ◽  
Author(s):  
D. F. Shanno
Keyword(s):  
Gradient Methods ◽  
Optimization Techniques ◽  
Newton Methods ◽  
Variable Metric Methods ◽  
Distributed Software ◽  
Variable Metric ◽  
Gradient Based ◽  
Truncated Newton Methods ◽  
Truncated Newton ◽  
Theoretical Results

The paper surveys recent results in conjugate gradient methods, variable storage variable metric methods, sparse variable metric and finite difference Newton methods, and truncated Newton methods. Both computational and theoretical results will be discussed, as well as currently distributed software.

Inverse Surfacelet Transform for Image Reconstruction With Constrained-Conjugate Gradient Methods

2014 ◽  
Vol 14 (2) ◽  
Author(s):  
Wei Huang ◽  
Yan Wang ◽  
David W. Rosen
Keyword(s):  
Image Reconstruction ◽  
Conjugate Gradient ◽  
Gradient Methods ◽  
Optimization Methods ◽  
Transformation Process ◽  
Geometric Constraints ◽  
Least Square ◽  
Reconstruction Methods ◽  
Gradient Based ◽  
Surfacelet Transform

Image reconstruction is the transformation process from a reduced-order representation to the original image pixel form. In materials characterization, it can be utilized as a method to retrieve material composition information. In our previous work, a surfacelet transform was developed to efficiently represent boundary information in material images with surfacelet coefficients. In this paper, new constrained-conjugate-gradient based image reconstruction methods are proposed as the inverse surfacelet transform. With geometric constraints on boundaries and internal distributions of materials, the proposed methods are able to reconstruct material images from surfacelet coefficients as either lossy or lossless compressions. The results between the proposed and other optimization methods for solving the least-square error inverse problems are compared.

scholarly journals Fast Path Planning for Autonomous Ships in Restricted Waters

2018 ◽  
Vol 8 (12) ◽  
pp. 2592 ◽  
Author(s):  
Hongguang Lyu ◽  
Yong Yin
Keyword(s):  
Path Planning ◽  
Potential Field ◽  
Gradient Methods ◽  
Artificial Potential Field ◽  
Time Step ◽  
Human Errors ◽  
International Regulations ◽  
Planning Algorithm ◽  
Gradient Based ◽  
Autonomous Ship

Presently, there is increasing interest in autonomous ships to reduce human errors and support intelligent navigation, where automatic collision avoidance and path planning is a key problem, especially in restricted waters. To solve this problem, a path-guided hybrid artificial potential field (PGHAPF) method is first proposed in this paper. It is essentially a reactive path-planning algorithm that provides fast feedback in a changeable environment, including dynamic target ships (TSs) and static obstacles, for steering an autonomous ship safely. The proposed strategy, which is a fusion of the potential field and gradient methods, consists of potential-based path planning for arbitrary static obstacles, gradient-based decision-making for dynamic TSs, and their combination with consideration of the prior path and waypoint selection optimization. A three-degree-of-freedom dynamic model of a Mariner class vessel and a low-level controller have been incorporated together in this method to ensure that the vessel’s positions are updated at each time step in order to acquire a more applicable and reliable trajectory. Simulations show that the PGHAPF method has the potential to rapidly generate adaptive, collision-free and International Regulations for Preventing Collisions at Sea (COLREGS)-constrained trajectories in restricted waters by deterministic calculations. Furthermore, this method has the potential to perform path planning on an electronic chart platform and to overcome some drawbacks of traditional artificial potential field (APF) methods.

scholarly journals Learning quantized neural nets by coarse gradient method for nonlinear classification

2021 ◽  
Vol 8 (3) ◽  
Author(s):  
Ziang Long ◽  
Penghang Yin ◽  
Jack Xin
Keyword(s):  
Neural Networks ◽  
Loss Function ◽  
Ad Hoc ◽  
Gradient Methods ◽  
Synthetic Data ◽  
Neural Nets ◽  
Theoretical Understanding ◽  
Performance Guarantees ◽  
Almost Everywhere ◽  
Gradient Based

AbstractQuantized or low-bit neural networks are attractive due to their inference efficiency. However, training deep neural networks with quantized activations involves minimizing a discontinuous and piecewise constant loss function. Such a loss function has zero gradient almost everywhere (a.e.), which makes the conventional gradient-based algorithms inapplicable. To this end, we study a novel class of biased first-order oracle, termed coarse gradient, for overcoming the vanished gradient issue. A coarse gradient is generated by replacing the a.e. zero derivative of quantized (i.e., staircase) ReLU activation composited in the chain rule with some heuristic proxy derivative called straight-through estimator (STE). Although having been widely used in training quantized networks empirically, fundamental questions like when and why the ad hoc STE trick works, still lack theoretical understanding. In this paper, we propose a class of STEs with certain monotonicity and consider their applications to the training of a two-linear-layer network with quantized activation functions for nonlinear multi-category classification. We establish performance guarantees for the proposed STEs by showing that the corresponding coarse gradient methods converge to the global minimum, which leads to a perfect classification. Lastly, we present experimental results on synthetic data as well as MNIST dataset to verify our theoretical findings and demonstrate the effectiveness of our proposed STEs.

Sign in / Sign up

Export Citation Format

Share Document