Lu Ren

AAAI Conference 2024 Conference Paper

Robustly Train Normalizing Flows via KL Divergence Regularization

• Kun Song
• Ruben Solozabal
• Hao Li
• Martin Takáč
• Lu Ren
• Fakhri Karray

In this paper, we find that the training of Normalizing Flows (NFs) are easily affected by the outliers and a small number (or high dimensionality) of training samples. To solve this problem, we propose a Kullback–Leibler (KL) divergence regularization on the Jacobian matrix of NFs. We prove that such regularization is equivalent to adding a set of samples whose covariance matrix is the identity matrix to the training set. Thus, it reduces the negative influence of the outliers and the small sample number on the estimation of the covariance matrix, simultaneously. Therefore, our regularization makes the training of NFs robust. Ultimately, we evaluate the performance of NFs on out-of-distribution (OoD) detection tasks. The excellent results obtained demonstrate the effectiveness of the proposed regularization term. For example, with the help of the proposed regularization, the OoD detection score increases at most 30% compared with the one without the regularization.

JMLR Journal 2011 Journal Article

Logistic Stick-Breaking Process

• Lu Ren
• Lan Du
• Lawrence Carin
• David Dunson

A logistic stick-breaking process (LSBP) is proposed for non-parametric clustering of general spatially- or temporally-dependent data, imposing the belief that proximate data are more likely to be clustered together. The sticks in the LSBP are realized via multiple logistic regression functions, with shrinkage priors employed to favor contiguous and spatially localized segments. The LSBP is also extended for the simultaneous processing of multiple data sets, yielding a hierarchical logistic stick-breaking process (H-LSBP). The model parameters (atoms) within the H-LSBP are shared across the multiple learning tasks. Efficient variational Bayesian inference is derived, and comparisons are made to related techniques in the literature. Experimental analysis is performed for audio waveforms and images, and it is demonstrated that for segmentation applications the LSBP yields generally homogeneous segments with sharp boundaries. [abs] [ pdf ][ bib ] &copy JMLR 2011. ( edit, beta )

NeurIPS Conference 2011 Conference Paper

The Kernel Beta Process

• Lu Ren
• Yingjian Wang
• Lawrence Carin
• David Dunson

A new Le ́vy process prior is proposed for an uncountable collection of covariate- dependent feature-learning measures; the model is called the kernel beta process (KBP). Available covariates are handled efficiently via the kernel construction, with covariates assumed observed with each data sample (“customer”), and latent covariates learned for each feature (“dish”). Each customer selects dishes from an infinite buffet, in a manner analogous to the beta process, with the added constraint that a customer first decides probabilistically whether to “consider” a dish, based on the distance in covariate space between the customer and dish. If a customer does consider a particular dish, that dish is then selected probabilistically as in the beta process. The beta process is recovered as a limiting case of the KBP. An efficient Gibbs sampler is developed for computations, and state-of-the-art results are presented for image processing and music analysis tasks.

NeurIPS Conference 2009 Conference Paper

A Bayesian Model for Simultaneous Image Clustering, Annotation and Object Segmentation

• Lan Du
• Lu Ren
• Lawrence Carin
• David Dunson

A non-parametric Bayesian model is proposed for processing multiple images. The analysis employs image features and, when present, the words associated with accompanying annotations. The model clusters the images into classes, and each image is segmented into a set of objects, also allowing the opportunity to assign a word to each object (localized labeling). Each object is assumed to be represented as a heterogeneous mix of components, with this realized via mixture models linking image features to object types. The number of image classes, number of object types, and the characteristics of the object-feature mixture models are inferred non-parametrically. To constitute spatially contiguous objects, a new logistic stick-breaking process is developed. Inference is performed efficiently via variational Bayesian analysis, with example results presented on two image databases.

IROS Conference 2009 Conference Paper

Hybrid control of door-opening by modular re-configurable Robots

• Guangjun Liu
• Saleh Ahmad
• Lu Ren

In this paper, we study the problem of door-opening by using modular reconfigurable robot (MRR). Based on multiple working mode control of the joint modules of the MRR, we first propose an online mode-switch strategy for all joint modules so that each joint module can be easily determined when it should be switched between active working mode and passive working mode during the door-opening process. Based on the proposed mode-switch strategy, a hybrid control scheme is proposed for door-opening. Simulation results are used to demonstrate the validity and efficiency of the proposed mode-switch strategy and the hybrid control scheme.

NeurIPS Conference 2009 Conference Paper

Non-Parametric Bayesian Dictionary Learning for Sparse Image Representations

• Mingyuan Zhou
• Haojun Chen
• Lu Ren
• Guillermo Sapiro
• Lawrence Carin
• John Paisley

Non-parametric Bayesian techniques are considered for learning dictionaries for sparse image representations, with applications in denoising, inpainting and compressive sensing (CS). The beta process is employed as a prior for learning the dictionary, and this non-parametric method naturally infers an appropriate dictionary size. The Dirichlet process and a probit stick-breaking process are also considered to exploit structure within an image. The proposed method can learn a sparse dictionary in situ; training images may be exploited if available, but they are not required. Further, the noise variance need not be known, and can be non-stationary. Another virtue of the proposed method is that sequential inference can be readily employed, thereby allowing scaling to large images. Several example results are presented, using both Gibbs and variational Bayesian inference, with comparisons to other state-of-the-art approaches.

ICML Conference 2008 Conference Paper

The dynamic hierarchical Dirichlet process

• Lu Ren
• David B. Dunson
• Lawrence Carin

IROS Conference 2007 Conference Paper

3-D automatic microassembly by vision-based control

• Lu Ren
• Lidai Wang
• James K. Mills
• Dong Sun 0001

In this paper, we propose a vision control strategy to perform automatic microassembly tasks in threedimension (3-D), and develop relevant control software. Specifically, using a 6 degree-of-freedom (DOF) robotic workstation to control a passive microgripper to automatically grasp a designated micropart from the chip, pivot the micropart, and then move the micropart to vertically insert into a designated slot on the chip. In the proposed control strategy, the whole microassembly task is divided into two subtasks, micro-grasping and micro-joining, in sequence. To guarantee the success of microassembly and manipulation accuracy, two different two-stage feedback motion strategies, the pattern matching and auto-focus method are employed, with the use of vision-based control system and the vision control software developed. Experiments conducted demonstrate the efficiency and validity of the proposed control strategy.

ICRA Conference 2006 Conference Paper

Convex Synchronized Control for a 3-DOF Planar Parallel Manipulator

• Lu Ren
• James K. Mills
• Dong Sun 0001

In this paper, in order to improve tracking accuracy and satisfy multiple closed-loop performance specifications simultaneously during high-speed, high-acceleration trajectory tracking for a 3 degree-of-freedom (DOF) planar parallel manipulator, we propose a new control approach, termed convex synchronized (C-S) control. The C-S control is base on the so-called convex combination method and synchronized control. To implement the synchronized control scheme, a feedback signal, termed the synchronization error, is employed, which represents the degree of coordination of the active joints in the parallel manipulator based on the kinematics of the parallel manipulator, and thus tracking accuracy is improved. On the other hand, the convex combination method utilizes the convex property of the required closed-loop performance specifications. Through combining multiple linear controllers, so-called sample controllers, each of which satisfies at least one closed-loop performance specification, a C-S controller is algebraically calculated, which satisfies all closed-loop performance specifications simultaneously. Compared with traditional trial-and-error method, the convex combination method is more straightforward and efficient. Hence, possessing advantages of the synchronized control and the convex combination method, the proposed C-S control method can both improve tracking accuracy and satisfy multiple closed-loop performance specifications simultaneously (MSS). Experiments conducted on a 3-DOF P-R-R type planar parallel manipulator demonstrate the above claims

ICRA Conference 2006 Conference Paper

Development of a 6 Degree of Freedom Robotic Micromanipulator for Use in 3D MEMS Microassembly

• Nikolai Dechev
• Lu Ren
• William Liu
• William L. Cleghorn
• James K. Mills

This paper describes the design and development of a 6 degree of freedom robotic manipulator used in the assembly of three-dimensional MEMS (micro electromechanical systems) microstructures. The robot employs a highly innovative mechanical design for the rotational axes to provide unprecedented access to a microchip substrate for microassembly operations. The first three axes of the robotic manipulator are orthogonally mounted linear stages providing Cartesian positioning of the chips beneath the end effector (microgripper). A rotational stage (alpha) mounted on the distal end of these three Cartesian axes allows the MEMS chip to be rotated. Two more degrees of freedom (beta and gamma) are serially mounted to the base frame, allowing for two degrees of rotation of the end effector. This configuration permits assembly of micro-parts on the surface of a MEMS chip at any orientation angle to the surface, within the limits of the workspace of the manipulator and the resolution of the motors. The end effector employs a standard tungsten probe with a passive microgripper bonded to it, which is used for grasping micro-parts. A software system has been developed to allow automatic operation of the manipulator. Preliminary assembly tests confirm the usefulness of the proposed design

IROS Conference 2006 Conference Paper

Performance Improvement of Tracking Control for a Planar Parallel Robot Using Synchronized Control

• Lu Ren
• James K. Mills
• Dong Sun 0001

In order to improve trajectory tracking accuracy for a three degree-of-freedom (DOF) planar parallel robot, in this paper, we develop a new control approach based on adaptive control with the use of the so-called synchronization error. Similar to the contour error proposed for machine tools, the defined synchronization error represents the degree of coordination amongst the active joints in the parallel robot, which is substantially different from the traditional tracking errors. By using the synchronization error, all active joints in the parallel robot are controlled to move in a synchronous manner so that the trajectory tracking accuracy of the robot end-effector is substantially improved. In addition, with the use of adaptive control, the synchronization error and the pose error of the platform are guaranteed to converge to zero simultaneously, while uncertain parameters in the system dynamic model are guaranteed to converge to their true values. Experiments conducted on the planar parallel robot verify the above claims and evaluate performance of the proposed control approach, compared with conventional PID control

ICRA Conference 2005 Conference Paper

Controller Design Applied to Planar Parallel Manipulators For Trajectory Tracking Control

• Lu Ren
• James K. Mills
• Dong Sun 0001

In this paper, we develop a new control method for a P-R-R type planar parallel manipulator, termed adaptive synchronized (A-S) control. The novelty of the proposed A-S control, a combination of synchronized control and adaptive control, is in the application of synchronized control to a planar parallel manipulator. To improve trajectory control, based on the kinematics of the planar parallel manipulator, we design a synchronization feedback control signal, termed the synchronization error, which represents the degree of coordination amongst the actuated joints. Employment of the synchronization error is shown to substantially reduce the pose error of the moving platform of the planar parallel manipulator during trajectory tracking. An adaptive controller is also used to estimate uncertain dynamic parameters of the manipulator. Under the assumption of persistent excitation, the proposed A-S control algorithm is theoretically proved to simultaneously guarantee the convergence of tracking errors and the synchronization error. Moreover, the estimated unknown parameters are guaranteed to converge to their true values as well. Finally, experiments are conducted to verify these claims and evaluate the performance of the proposed controller. Experimental results show that A-S control yields good trajectory tracking performance.

ICRA Conference 2005 Conference Paper

Nonlinear PD Synchronized Control for Parallel Manipulators

• Yuxin Su 0002
• Dong Sun 0001
• Lu Ren
• Xiaoyun Wang 0003
• James K. Mills

A simple synchronized control algorithm is proposed, by incorporating cross-coupling technology into a common PD control architecture, for control of parallel manipulators. A saturated proportional (S-P) control and a linear proportional derivative (PD) control plus gravity force compensation is implemented for synchronization and position control, respectively. The proposed control law is easy to implement and is able to stabilize motion of each actuator while synchronizing all actuators’ motions, so that differential position errors amongst actuators converge to zero. It is shown that the proposed method guarantees global asymptotical stability of the closed-loop system. Experiments performed on a three-DOF parallel manipulator demonstrate the effectiveness of the proposed approach.