Unsupervised Learning for Maximum Consensus Robust Fitting: A Reinforcement Learning Approach

Robust model fitting is a core algorithm in several computer vision applications. Despite being studied for decades, solving this problem efficiently for datasets that are heavily contaminated by outliers is still challenging: due to the underlying computational complexity. A recent focus has been on learning-based algorithms. However, most of these approaches are supervised (which require a large amount of labelled training data). In this paper, we introduce a novel unsupervised learning framework: that learns to directly (without labelled data) solve robust model fitting. Moreover, unlike other learning-based methods, our work is agnostic to the underlying input features, and can be easily generalized to a wide variety of LP-type problems with quasi-convex residuals. We empirically show that our method outperforms existing (un)supervised learning approaches, and also achieves competitive results compared to traditional (non-learning-based) methods. Our approach is designed to try to maximise consensus (MaxCon), similar to the popular RANSAC. The basis of our approach, is to adopt a Reinforcement Learning framework. This requires designing appropriate reward functions, and state encodings. We provide a family of reward functions, tunable by choice of a parameter. We also investigate the application of different basic and enhanced Q-learning components.

A pilot study on the biochemical effects of repeated administration of 24% oral sucrose vs. 30% oral dextrose on urinary markers of adenosine triphosphate degradation

OBJECTIVES

Premature neonates often receive oral sucrose or dextrose before tissue-damaging procedures (TDPs). Previous work showed that a single dose of sucrose, but not dextrose, increased cellular energy utilization and ATP degradation. This pilot study probes the effects of repeated administration of sucrose or dextrose on energy metabolism.

METHODS

Urinary markers of ATP metabolism (hypoxanthine, xanthine, uric acid) are measured in premature neonates randomized to receive: (a) standard of care, (b) 0.2 ml 24% oral sucrose, or (c) 0.2 ml 30% oral dextrose, before every painful procedure on days-of-life 3-7.

RESULTS

Standard of care is associated with highest xanthine/creatinine and uric acid/creatinine, likely because of fewer pain treatments. Benefits of repeated oral sucrose are unclear. Neonates receiving oral dextrose had lower xanthine/creatinine and uric acid/creatinine.

CONCLUSIONS

Repeated treatments of neonatal procedural pain with 30% oral dextrose are less energetically demanding. Larger clinical studies are needed for comparison with sucrose treatments.

Oral dextrose reduced procedural pain without altering cellular ATP metabolism in preterm neonates: a prospective randomized trial

OBJECTIVE

To examine the effects of 30% oral dextrose on biochemical markers of pain, adenosine triphosphate (ATP) degradation, and oxidative stress in preterm neonates experiencing a clinically required heel lance.

STUDY DESIGN

Utilizing a prospective study design, preterm neonates that met study criteria (n = 169) were randomized to receive either (1) 30% oral dextrose, (2) facilitated tucking, or (3) 30% oral dextrose and facilitated tucking 2 min before heel lance. Plasma markers of ATP degradation (hypoxanthine, uric acid) and oxidative stress (allantoin) were measured before and after the heel lance. Pain was measured using the premature infant pain profile-revised (PIPP-R).

RESULTS

Oral dextrose, administered alone or with facilitated tucking, did not alter plasma markers of ATP utilization and oxidative stress.

CONCLUSION

A single dose of 30% oral dextrose, given before a clinically required heel lance, decreased signs of pain without increasing ATP utilization and oxidative stress in premature neonates.

Association of Umbilical Cord Milking vs Delayed Umbilical Cord Clamping With Death or Severe Intraventricular Hemorrhage Among Preterm Infants

IMPORTANCE

Umbilical cord milking as an alternative to delayed umbilical cord clamping may provide equivalent benefits to preterm infants, but without delaying resuscitation.

OBJECTIVE

To determine whether the rates of death or severe intraventricular hemorrhage differ among preterm infants receiving placental transfusion with umbilical cord milking vs delayed umbilical cord clamping.

DESIGN, SETTING, AND PARTICIPANTS

Noninferiority randomized clinical trial of preterm infants (born at 23-31 weeks' gestation) from 9 university and private medical centers in 4 countries were recruited and enrolled between June 2017 and September 2018. Planned enrollment was 750 per group. However, a safety signal comprising an imbalance in the number of severe intraventricular hemorrhage events by study group was observed at the first interim analysis; enrollment was stopped based on recommendations from the data and safety monitoring board. The planned noninferiority analysis could not be conducted and a post hoc comparison was performed instead. Final date of follow-up was December 2018.

INTERVENTIONS

Participants were randomized to umbilical cord milking (n = 236) or delayed umbilical cord clamping (n = 238).

MAIN OUTCOMES AND MEASURES

The primary outcome was a composite of death or severe intraventricular hemorrhage to determine noninferiority of umbilical cord milking with a 1% noninferiority margin.

RESULTS

Among 540 infants randomized, 474 (88%) were enrolled and completed the trial (mean gestational age of 28 weeks; 46% female). Twelve percent (29/236) of the umbilical cord milking group died or developed severe intraventricular hemorrhage compared with 8% (20/238) of the delayed umbilical cord clamping group (risk difference, 4% [95% CI, -2% to 9%]; P = .16). Although there was no statistically significant difference in death, severe intraventricular hemorrhage was statistically significantly higher in the umbilical cord milking group than in the delayed umbilical cord clamping group (8% [20/236] vs 3% [8/238], respectively; risk difference, 5% [95% CI, 1% to 9%]; P = .02). The test for interaction between gestational age strata and treatment group was significant for severe intraventricular hemorrhage only (P = .003); among infants born at 23 to 27 weeks' gestation, severe intraventricular hemorrhage was statistically significantly higher with umbilical cord milking than with delayed umbilical cord clamping (22% [20/93] vs 6% [5/89], respectively; risk difference, 16% [95% CI, 6% to 26%]; P = .002).

CONCLUSIONS AND RELEVANCE

In this post hoc analysis of a prematurely terminated randomized clinical trial of umbilical cord milking vs delayed umbilical cord clamping among preterm infants born at less than 32 weeks' gestation, there was no statistically significant difference in the rate of a composite outcome of death or severe intraventricular hemorrhage, but there was a statistically significantly higher rate of severe intraventricular hemorrhage in the umbilical cord milking group. The early study termination and resulting post hoc nature of the analyses preclude definitive conclusions.

TRIAL REGISTRATION

ClinicalTrials.gov Identifier: NCT03019367.

A Randomized Clinical Trial of Umbilical Cord Milking vs Delayed Cord Clamping in Preterm Infants: Neurodevelopmental Outcomes at 22-26 Months of Corrected Age

OBJECTIVE

To compare the effect of umbilical cord milking vs delayed cord clamping (DCC) on neurodevelopmental and health outcomes in very preterm infants at 22-26 months of corrected age.

STUDY DESIGN

Neurodevelopmental outcomes at 2 years of age were assessed using the Bayley Scales of Infant Development, third edition, and a standardized neurologic examination. Data regarding pulmonary morbidities, neurosensory impairments, and hospitalizations were obtained by parental interview. Intention-to-treat was used for primary analyses.

RESULTS

Of the 197 infants enrolled in the original study there were 15 deaths, 5 in the umbilical cord milking group and 10 in DCC group. Of the remaining infants, 135 (74%) were assessed at 22-26 months of corrected age. Demographics in umbilical cord milking (n = 70) and DCC (n = 65) groups were similar. Infants randomized to umbilical cord milking at birth had significantly higher cognitive and language composite scores, and were less likely to have a cognitive composite score of <85 (4% vs 15%; P = .04). Motor function was similar in both groups. There were no differences in the incidences of mild or moderate to severe neurodevelopmental impairment, hearing or visual impairments, pulmonary morbidities, or rehospitalizations between the 2 groups.

CONCLUSIONS

Infants randomized to umbilical cord milking had higher language and cognitive scores compared with those randomized to DCC. There was no difference in rates of mild or moderate to severe neurodevelopmental impairment.

TRIAL REGISTRATION

clinicaltrials.gov: NCT01434732.

Umbilical Cord Milking Versus Delayed Cord Clamping in Preterm Infants

BACKGROUND AND OBJECTIVE

Delayed cord clamping (DCC) is recommended for premature infants to improve blood volume. Most preterm infants are born by cesarean delivery (CD), and placental transfusion may be less effective than in vaginal delivery (VD). We sought to determine whether infants <32 weeks born by CD who undergo umbilical cord milking (UCM) have higher measures of systemic blood flow than infants who undergo DCC.

METHODS

This was a 2-center trial. Infants delivered by CD were randomly assigned to undergo UCM or DCC. Infants delivered by VD were also randomly assigned separately. UCM (4 strippings) or DCC (45-60 seconds) were performed. Continuous hemodynamic measurements and echocardiography were done at site 1.

RESULTS

A total of 197 infants were enrolled (mean gestational age 28 ± 2 weeks). Of the 154 infants delivered by CD, 75 were assigned to UCM and 79 to DCC. Of the infants delivered by CD, neonates randomly assigned to UCM had higher superior vena cava flow and right ventricular output in the first 12 hours of life. Neonates undergoing UCM also had higher hemoglobin, delivery room temperature, blood pressure over the first 15 hours, and urine output in the first 24 hours of life. There were no differences for the 43 infants delivered by VD.

CONCLUSIONS

This is the first randomized controlled trial demonstrating higher systemic blood flow with UCM in preterm neonates compared with DCC. UCM may be a more efficient technique to improve blood volume in premature infants delivered by CD.

A comparison of commonly used and novel electronic techniques for evaluating cattle temperament

Schwartzkopf-Genswein, K. S., Shah, M. A., Church, J. S., Haley, D. B., Janzen, K., Truong, G., Atkins, R. P. and Crowe, T. J. 2012. A comparison of commonly used and novel electronic techniques for evaluating cattle temperament. Can. J. Anim. Sci. 92: 21–31. The temperament of steers (n=28) was assessed using five quantitative techniques including: flight time, flight distance, electronic (strain-gauge and accelerometer) tests, and three visual scores (VS) made during entry, restraint and exit from a squeeze chute. The objective of this study was to determine the most important predictive parameters based on those measurements and evaluate the relationship between the techniques. Flight time and distance were correlated with exit VS (r=−0.51, and 0.41, P<0.05; n=56), but were not related to restraint VS. Data from strain-gauge and accelerometer sensors were used to generate parameters such as peak response and area under the curve that were correlated with all three VS. Regression models using VS as the dependent variable and a combination of 2 to 5 parameters from the strain-gauge and accelerometer tests as independent variables predicted temperament with values of 29 to 65 or 41 to 57%, respectively. When all techniques, excluding VS, were used as independent variables, model accuracy increased to 72, 81 and 77% for restraint, exit and the sum of all VS, respectively. These findings suggest the objective measures of temperament assessed in this study could be used to identify highly reactive animals.

Electron spin control in dilute nitride semiconductors

We report on a study of spin-dependent recombination processes (SDR) for conduction band electrons on deep paramagnetic centers in a series of GaAs(1-y)N(y) epilayers by time-resolved optical orientation experiments. We demonstrate that this dilute nitride compound can be used as an effective electron spin filter under a polarized optical excitation of appropriate intensity. This optimum intensity can moreover be controlled by adjusting the nitrogen composition in the layer.

Diode-pumped passively mode-locked Nd:YAG laser with 10-W average power in a diffraction-limited beam

We present a passively mode-locked Nd:YAG laser with 10.7-W average output power in a diffraction-limited beam. Stable self-starting mode locking with a pulse duration of 16 ps and a pulse energy of 120 nJ is obtained with a semiconductor saturable-absorber mirror. The laser is directly side pumped with two 20-W diode bars. Single-pass frequency doubling in an external 5-mm-thick KTP crystal yields 3.2-W average power at 532 nm.