Data-driven algorithm selection and tuning in optimization and signal processing

Machine learning algorithms typically rely on optimization subroutines and are well known to provide very effective outcomes for many types of problems. Here, we flip the reliance and ask the reverse question: can machine learning algorithms lead to more effective outcomes for optimization problems? Our goal is to train machine learning methods to automatically improve the performance of optimization and signal processing algorithms. As a proof of concept, we use our approach to improve two popular data processing subroutines in data science: stochastic gradient descent and greedy methods in compressed sensing. We provide experimental results that demonstrate the answer is “yes”, machine learning algorithms do lead to more effective outcomes for optimization problems, and show the future potential for this research direction. In addition to our experimental work, we prove relevant Probably Approximately Correct (PAC) learning theorems for our problems of interest. More precisely, we show that there exists a learning algorithm that, with high probability, will select the algorithm that optimizes the average performance on an input set of problem instances with a given distribution.

     

Infants entering foster care compared to other infants using birth status indicators

OBJECTIVE: Infants comprise nearly one fourth of all entries to foster care. Linkage of administrative birth and placement data can provide information about these infants that may otherwise be unavailable or difficult to obtain. METHOD: Statewide birth records and foster care placement histories were linked via probability matching. Logit regression was used to compare 26,460 maltreated infants who entered foster care between 1989 and 1994 with a random sample of 68,401 other infants born during that time frame. RESULTS: Infants in care were more than twice as likely to have single parents and be born with low birthweight, and twice as likely to have been born with a birth abnormality as other infants, controlling for other factors. The largest difference was in the eightfold increased likelihood for mothers of infants in care to have had no prenatal care. Infants in care were nearly three times as likely to be born into larger families (third or greater live births to the mother). Mothers of infants in care were more than twice as likely to be African American compared to White than mothers of other infants, while Hispanic and Other ethnic groups were underrepresented in the group of infants in care. Foreign born mothers, especially Hispanic women, were much less likely to have infants in care than they were to have children in the other group. CONCLUSIONS: Administrative datasets, while often limited in the number of variables they include and scope of their information, can be a valuable tool when used to understand demographics and frame questions for future research. Infants who enter foster care differ in substantial ways from other children. These findings have important implications for future research aimed toward targeting of child welfare services and supports.     

Comparison of fat-saturation fast spin echo versus conventional spin-echo MRI in the detection of rotator cuff pathology

The purpose of this study was to compare the diagnostic performance of fat-saturation fast-spin-echo (FSE) T2-weighted (T2W) sequences with conventional spin-echo (CSE) T2W sequences in the detection of rotator cuff pathology using surgery as the reference standard. Oblique coronal dual-echo CSE and FSE T2W images with fat saturation from 50 surgically confirmed MR shoulder examinations were acquired on a 1.5-T MR scanner. Blinded MR readers retrospectively analyzed each imaging sequence separately and ultimately correlated both sequences together with findings at surgery. FSE was 100% sensitive and 94% specific in detection of full-thickness tears (n = 19) and 73% sensitive and 97% specific in the detection of partial-thickness rotator cuff tears (n = 13). There was no statistically significant difference in the performance of FSE with fat saturation compared with CSE. The two discrepancies between imaging sequences related to the extent of partial-thickness tears. Our findings suggest that fat-saturation FSE imaging can effectively replace CSE imaging in the evaluation of rotator cuff pathology.