Search Results (1 - 3 of 3 Results)

Sort By  
Sort Dir
 
Results per page  

Ewing, GabrielKnowledge Transfer from Expert Demonstrations in Continuous State-Action Spaces
Master of Sciences, Case Western Reserve University, 2018, EECS - Computer and Information Sciences
In this thesis, we address the task of reinforcement learning in continuous state and action spaces. Specifically, we consider multi-task reinforcement learning, where a sequence of reinforcement learning tasks have to be solved, and inverse reinforcement learning, where a reward function has to be learned from expert demonstrations. We also use trees to represent models, rewards, and value functions in our domains. First, we design an algorithm to learn from demonstration in the presence of a non- smooth reward function. Second, we design another algorithm to perform offline reinforcement learning in the same scenario. This allows us to re-use experiences to help with new tasks. Third, we introduce a method to incorporate weak knowledge about policies with online learning in policy gradient algorithms. These contributions allow us to create a pipeline that efficiently learns and transfers knowledge across a sequence of tasks. We demonstrate our approaches on the task of learning control of a prosthetic arm from expert demonstrations under various scenarios in simulation.

Committee:

Soumya Ray, Dr. (Advisor); Michael Fu, Dr. (Committee Member); M. Cenk Cavusoglu, Dr. (Committee Member); Michael Lewicki, Dr. (Committee Member)

Subjects:

Computer Science

Keywords:

Machine learning; reinforcement learning; continuous actions; knowledge transfer; prostheses

Hu, XingxueAnalyses of Effects of Pigments on Maxillofacial Prosthetic Material
Doctor of Philosophy, The Ohio State University, 2010, Oral Biology

Fabrication of a successful maxillofacial prosthesis with appearance match to the patient’s skin is often a significant challenge. An accurate method for collecting appearance information and the application of advanced colorant formulation theories would make possible a pleasing appearance match of a maxillofacial prosthesis to the patient’s skin.

Dental and maxillofacial applications of a non-contact spectral reflectance system with its potential to provide more reliable measurements recently initiated. A study on accuracy and precision of this measuring system is conducted by comparing with four commonly used contact color measuring instruments, which provides more evidence for its further application on color matching of maxillofacial prosthesis.

A newly-developed laser light diffusing method, which correlates laser light diffusing area (LLDA) to translucency parameter (TP), is investigated on maxillofacial elastomeric specimens by comparing a previously proposed method, color difference due to edge loss (CDEL). The translucency of human skin and translucent maxillofacial materials can be described quantitatively with this non-destructive, non-contact method, which might be further applied on application of reproduction of facial morphology and appearance.

Pigments and dyes play a key role in pigmentation and coloration of maxillofacial prosthetic elastomers. Addition of pigments and dyes into the base material might produce complicated reactions or interactions which in turn may cause various effects on optical properties, biocompatibility and thermal properties of the material.

Optical properties of a base maxillofacial prosthetic material are significantly affected when pigmentation by pigments and dyes. Kubelka-Munk (K-M) theory has been widely used on pigmentation and coloration of maxillofacial materials. Accuracy of the K-M theory with three different interfacial reflection corrections (IRC) on maxillofacial elastomers is evaluated, and the IRC value for translucent materials provides the least error for pigmented maxillofacial prosthetic elastomers. Furthermore, rather than the first order linear regression model traditionally used in concentration additivity, a newly proposed second order regression model in concentration additivity is evaluated and recognized as a regression model with the least error in colorant formulation based on the K-M theory for pigmented maxillofacial materials.

Biocompatibility of maxillofacial prosthetic elastomer is potentially affected after pigmentation and coloration. Cytotoxicity testing is used to initially investigate effects of pigments on biocompatibility of the material. Indirect test with olorimetric assay of MTS (an MTT analog), and direct test with digital imaging are performed. Maxillofacial prosthetic materials show a minor cytotoxicity, but no practically significant effects are found on cytotoxicity of pigmented maxillofacial elastomers.

Differential scanning calorimetric analysis is used to investigate effects of pigments on thermal properties of pigmented maxillofacial elastomers. A strong endothermic peak associated with melting and recrystalization is investigated and found to be at approximately -43°C, with an enthalpy change of approximately 11 J/g, which is consistent with results found in previous studies. No practically significant effects of pigments are found on pigmented maxillofacial elastomers, suggesting no significant effects on mechanical properties of the materials as well.

Committee:

William Johnston (Advisor); William Brantley (Committee Member); Robert Seghi (Committee Member); John Walters (Committee Member)

Subjects:

Biomedical Research; Dental Care; Health Care; Materials Science; Physics; Polymers

Keywords:

pigments; maxillofacial elastomer;maxillofacial prostheses; color; translucency; reflectance theory; cytotoxicity; differential scanning calorimetry;

Lertmanorat, ZengAn Electrode Array for Reversing the Recruitment Order of Peripheral Stimulation
Doctor of Philosophy, Case Western Reserve University, 2004, Biomedical Engineering

Extracellular electrical stimulation of peripheral nerve activates large diameter fibers before small ones. Recruitment order from small to large diameters is more physiological and desirable in many applications. Current methods to solve this problem require long duration stimulus pulses leading to electrode corrosion. The hypothesis that selective activation of small axons can be achieved by reshaping the extracellular potential profile along axons using an array of electrodes was tested in computer simulations and animal experiments.

Simulations in homogenous medium using an array of 4 cathodes and 5 anodes with 0.75mm contact separation showed that large axons (13-17µm) (0.6-3.0mA) had higher stimulating threshold than small axons (2-7µm) (0.4-0.7mA) with 200µm axon-electrode distance and 10µs stimulus pulse. In finite element model of dog sacral root where the fiber diameter distribution was bimodal with peaks at 5 and 14 µm (389 axons), 5 and 7-contact arrays recruited 35% and 65% of small axons (<10 µm) when recruiting only 10% and 30% of larger axons. Effectiveness of 9 and 11-contact arrays was decreased with the presence of epineurium and perineurium. Pulsewidth of monophasic stimulation had no effect on the diameter selective stimulation of the array. For the biphasic pulse stimulation, amplitude of the second phase has to be set lower than the first phase to prevent the excitation of large axons.

Effect of electrode arrays of 5, 7, and 11 contacts on the recruitment order were tested in computer simulations and animal experiments using cat Lateral Gastrocnemius (LG) innervating both LG (10-14µm) and soleus (8-12µm) muscle groups. The results in computer simulations and animal experiments were consistent. Tripolar electrode activated LG before soleus, whereas the 5-contact array shifted the soleus recruitment curve to overlap with that of LG, and the 7 and 11 contacts reversed the recruitment order between LG and soleus. Effect of the 11-contact array was not highly reproducible experimentally. The arrays decreased the recruitment curve slope to only 10% that of tripolar electrode. The 5 and 7-contact array can be used to reverse the recruitment order of peripheral nerve stimulation.

Committee:

Dominique Durand (Advisor)

Subjects:

Engineering, Biomedical

Keywords:

neural prostheses; recruitment order; Selective nerve stimulation