Multiclass classification with bandit feedback using adaptive regularization

We present a new multiclass algorithm in the bandit framework, where after making a prediction, the learning algorithm receives only partial feedback, i.e., a single bit indicating whether the predicted label is correct or not, rather than the true label. Our algorithm is based on the second-order Perceptron, and uses upper-confidence bounds to trade-off exploration and exploitation, instead of random sampling as performed by most current algorithms. We analyze this algorithm in a partial adversarial setting, where instances are chosen adversarially, while the labels are chosen according to a linear probabilistic model which is also chosen adversarially. We show a regret of $\mathcal{O}(\sqrt{T}\log T)$ , which improves over the current best bounds of $\mathcal{O}(T^{2/3})$ in the fully adversarial setting. We evaluate our algorithm on nine real-world text classification problems and on four vowel recognition tasks, often obtaining state-of-the-art results, even compared with non-bandit online algorithms, especially when label noise is introduced.     

Cognitive screening examinations for silent cerebral infarcts in sickle cell disease

OBJECTIVE: In children with sickle cell disease (SCD), silent cerebral infarcts are the most frequent cause of neurologic injury. We determined the sensitivity and specificity of selective neurocognitive measures when separating children with silent cerebral infarcts and SCD from sibling controls. Additionally, we tested the validity of the same cognitive measures to identify patients with overt strokes. METHODS: We examined performance on a neuropsychologic battery containing measures of attention/executive, spatial, language, memory, and motor functioning for seven children with SCD and silent cerebral infarct, 21 children with SCD and overt stroke, and 17 normal siblings. Diagnosis of cerebral infarct was based on results of MRI. RESULTS: Measures from the attention and executive domains were the most useful for identifying children with silent cerebral infarct. The Test of Variables of Attention was the most robust measure and yielded a sensitivity rate of 86% and a specificity rate of 81%. This measure also showed a sensitivity rate of 95% in identifying overt stroke. CONCLUSIONS: Brief cognitive screening measures, if properly constructed, may be an effective means of identifying children with silent cerebral infarct. Future prospective studies should be pursued to assess the utility of cognitive screening for silent cerebral infarcts in SCD.     

Proteolysis of the phage lambda CII regulatory protein by FtsH (HflB) of Escherichia coli

BACKGROUND: It has been suggested that the expression of psychosis may reflect an active morbid process that is associated with increasingly poor outcome unless ameliorated by antipsychotic drugs. METHODS: The subjects of this study were 48 in-patients with schizophrenia, many of whom had been admitted before the introduction of antipsychotic drugs to rural Irish psychiatric hospitals in the late 1950s. Each patient was assessed for positive and negative symptoms, and for general and executive (frontal) cognitive function. RESULTS: After controlling for age and for duration and continuity of subsequent antipsychotic treatment, current severity both of negative symptoms and of general cognitive impairment was predicted strongly by increasing duration of initially untreated psychosis; duration of illness following initiation of antipsychotic medication failed to predict the severity thereof. Neither of these indices of illness duration predicted the severity of positive symptoms or of executive dyscontrol. CONCLUSIONS: Increasing duration of initially untreated psychosis was associated specifically with heightened accrual of prominent negative symptoms and general cognitive impairment. Executive dyscontrol, though also prominent in these patients, may be 'locked-in' at an earlier phase of the illness.     

The effect of circulating fluidized bed boiler load on the emission of mercury

The paper deals with the determination of mercury balance for a hard coal-fired circulating fluidized bed boiler operated at various loads (i.e. 100%, 75% and 50% of the Maximum continuous rating, MCR). The research was focused on the determination of the mercury balance by providing data on Hg content in fuel, limestone sorbent, and in the combustion by-products (fly ash and bottom ash). Furthermore, the concentration of mercury in the flue gas in two zones i.e. upstream and downstream the electrostatic precipitator (ESP) were also determined in order to assess the effectiveness of the ESP in mercury reduction. For the calculation of the Hg balance both gaseous mercury species, Hg⁰, and oxidized mercury, Hg²⁺, were taken into consideration.The results indicated that over 95% of the mercury was fed into the boiler system with coal and just less than 5% with limestone used for ‘dry’ desulfurization of the flue gas. As for the boiler output, very high mercury content was determined in the fly ashes, while just a minor part (a few percent of the total mercury input) was found in the bottom ashes. The shares of Hg⁰ and Hg²⁺ mercury were affected by boiler load.     

On the Learnability and Design of Output Codes for Multiclass Problems

Output coding is a general framework for solving multiclass categorization problems. Previous research on output codes has focused on building multiclass machines given predefined output codes. In this paper we discuss for the first time the problem of designing output codes for multiclass problems. For the design problem of discrete codes, which have been used extensively in previous works, we present mostly negative results. We then introduce the notion of continuous codes and cast the design problem of continuous codes as a constrained optimization problem. We describe three optimization problems corresponding to three different norms of the code matrix. Interestingly, for the l ₂ norm our formalism results in a quadratic program whose dual does not depend on the length of the code. A special case of our formalism provides a multiclass scheme for building support vector machines which can be solved efficiently. We give a time and space efficient algorithm for solving the quadratic program. We describe preliminary experiments with synthetic data show that our algorithm is often two orders of magnitude faster than standard quadratic programming packages. We conclude with the generalization properties of the algorithm.     

Change in diurnal temperature rhythm in manic-depressive illness

The temperatures of six manic-depressive patients were taken every three hours consecutively for many weeks, covering at least one depressive and one manic episode in each patient. While the daily temperature curve was essentially normal in manic phases, with pronounced 24-hour rhythm, during depression the daytime temperatures appeared disorganised, often falling during the morning instead of rising, and with suggestions of a 12-hour rhythm. It may be useful to look on manic-depressive illness as resulting from a desynchronisation of circadian rhythms and to compare the pharmacologies of temperature regulation and mood regulation in psychosis.     

Online ranking by projecting

We discuss the problem of ranking instances. In our framework, each instance is associated with a rank or a rating, which is an integer in 1 to k. Our goal is to find a rank-prediction rule that assigns each instance a rank that is as close as possible to the instance's true rank. We discuss a group of closely related online algorithms, analyze their performance in the mistake-bound model, and prove their correctness. We describe two sets of experiments, with synthetic data and with the EachMovie data set for collaborative filtering. In the experiments we performed, our algorithms outperform online algorithms for regression and classification applied to ranking.     

The potential of Velvet bean (Mucuna pruriens) and N fertilizers in maize production on contrasting soils and agro-ecological zones of East Uganda

Research was conducted at two sites located in medium and low altitude zones in eastern Uganda. The aim of the study was to evaluate the benefit of Velvet bean (Mucuna pruriens) and inorganic N fertilizer in improving maize production in contrasting agro-ecological zones over two seasons. The medium altitude zone (Bulegeni) is a high-potential agricultural zone, with much more reliable rainfall and soils with high-productivity rating. The opposite is true for the low-altitude zone (Kibale). The soils were fertile for the site in the high-potential zone and poor in the low-potential zone. Over 22 weeks of fallow or relay with maize, Mucuna produced on average 8.2 t ha^–1 dry matter, accumulating 170 kg N ha^–1, with 57% of the N derived from the atmosphere in the low-potential zone, compared to 11.6 t ha^–1 dry matter, 350 kg N ha^–1, with 43% of the N derived from air, in the high-potential zone. Between 77 and 97% of the Mucuna-accumulated N was released over a period of 25 weeks, at a rate of 0.081 and 0.118 week^–1 in the high- and low-potential zones, respectively. The N-balance study shows that 93% of the applied N was accounted for in the high-potential zone, compared to 61% in the low-potential zone, due to differences in soil texture, soil fertility and maize biomass production at the two sites. As much as 44–73% of the N remained in the soil in the high-potential zone, compared to 39–53% in the low-potential zone, which might benefit the subsequent crops. There was a significant increase in maize yield in response to the added N, both from urea or Mucuna. The average increment above the control (continuous maize) was 3.2 t ha^–1 in the high-potential zone and 1.0 t ha^–1 in the low-potential zone. The maize yield increase over two seasons added up to 3.1 t ha^–1 with the application of inorganic fertilizers, and 1.9 t ha^–1 with a preceding Mucuna–maize relay in the high-potential zone, compared to an average of, 1.7 t ha^–1 with application of inorganic fertilizers and with Mucuna–maize relay in the low-potential zone. Application of P fertilizers with either N supply strategy significantly increased maize yield in the low-potential zone only, resulting in an additional 0.8 t ha^–1 for the inorganic N fertilizers and 1.3 t ha^–1 for a preceding Mucuna–maize relay. Apparently, P fertilizers are needed on poor soils. Clearly farmers stand to gain in terms of maize production from fertilizers as well as from the use of Mucuna, with more benefits from inorganic fertilizers in the high-potential zone.