Search

A distribution shift between the training and test data can severely harm performance of machine learning models. Importance weighting addresses this issue by assigning different weights to data points during training. We argue that existing heuristics for determining the weights are suboptimal, as they neglect the increase of the variance of the estimated model due to the finite sample size of the training data. We interpret the optimal weights in terms of a bias-variance trade-off, and propose a bi-level optimization procedure in which the weights and model parameters are optimized simultaneously. We apply this optimization to existing importance weighting techniques for last-layer retraining of deep neural networks in the presence of sub-population shifts and show empirically that optimizing weights significantly improves generalization performance., Comment: Preprint. Currently under review

Out-of-distribution generalization in neural networks is often hampered by spurious correlations. A common strategy is to mitigate this by removing spurious concepts from the neural network representation of the data. Existing concept-removal methods tend to be overzealous by inadvertently eliminating features associated with the main task of the model, thereby harming model performance. We propose an iterative algorithm that separates spurious from main-task concepts by jointly identifying two low-dimensional orthogonal subspaces in the neural network representation. We evaluate the algorithm on benchmark datasets for computer vision (Waterbirds, CelebA) and natural language processing (MultiNLI), and show that it outperforms existing concept removal methods, Comment: Preprint. Under Review. 33 pages

Phosphorus is an essential element for life, and the phosphorous cycle is widely believed to be a key factor limiting the extent of Earth's biosphere and its impact on remotely detectable features of Earth's atmospheric chemistry. Continental weathering is conventionally considered to be the only source of bioavailable phosphorus to the marine biosphere, with submarine hydrothermal processes acting as a phosphorus sink. Here, we use a novel 29Si tracer technique to demonstrate that alteration of submarine basalt under anoxic conditions leads to significant soluble phosphorus release, with an estimated ratio between phosphorus release and CO2 consumption (P/CO2) of 3.99+/-1.03 umol/mmol. This ratio is comparable to that of modern rivers, suggesting that submarine weathering under anoxic conditions is potentially a significant source of bioavailable phosphorus to planetary oceans and that volatile-rich Earth-like planets lacking exposed continents could develop robust biospheres capable of sustaining remotely detectable atmospheric biosignatures., Comment: Manuscript under consideration in Geophysical Research Letters

Different kinds of "gender gap" have been reported in different walks of the scientific life, almost always favouring male scientists over females. In this work, for the first time, we present a large-scale empirical analysis to ask whether female scientists with the same level of scientific accomplishment are as likely as males to be recognised. We particularly focus on Wikipedia, the open online encyclopedia that its open nature allows us to have a proxy of community recognition. We calculate the probability of appearing on Wikipedia as a scientist for both male and female scholars in three different fields. We find that women in Physics, Economics and Philosophy are considerable less likely than men to be recognised on Wikipedia across all levels of achievement., Comment: Under review

Research on descending motor pathways to caudal brainstem and spinal cord in the spinal motor system is reviewed. Particular attention is given to somatic and autonomic motoneurons in the spinal cord and brainstem, local projections to motoneurons, bulbospinal interneurons projecting to motoneurons, descending pathways of somatic motor control systems, and descending pathways involved in limbic motor control systems.

The descending pathways to the motoneuronal cell group of the cutaneous trunci muscle (CTM) of the cat were investigated by injecting H-3-labeled lucine into the brain stem, the diencephalon, or the C1, C2, C6, and C8 segments of the spinal cord, and examining fixed autoradiographic sections of the spinal cord and brain regions. Results demonstrate presence of specific supraspinal projectons to the CTM motor nucleus originating in the contralateral nucleus retroambiguous and the ipsilateral dorsolateral pontine tegmentum. Results also suggest that propriospinal pathways to the CTM motor nucleus originating in the cervical cord do not exist, although these propriospinal projections to all other motoneuronal cell groups surrounding the CTM nucleus are very strong.

Horseradish peroxidase was injected, or implanted unilaterally, into various levels of the spinal cord of anesthetized cats, to trace the distribution of projections to the spinal cord, of neurons in Field H of Forel, including the rostral interstitial nucleus of the medial longitudinal fasciculus (riMLF), and the interstitial nucleus of Cajal with adjacent reticular formation (INC-RF). Results indicate that, unlike the neurons projecting to the extraocular muscle motoneurons, the major portion of the spinally projecting neurons are not located in the riMLF or INC proper, but in adjacent areas, i.e., the ventral and lateral parts of the caudal third of the Field H of Forel and in the INC-RF. Neurons in caudal Field H of Forel, project, via the ventral part of the ventral funicululs, to the lateral part of the upper cervical ventral horn.

Results are presented of an anatomical study of the neuronal pathways in the cat, via which the periaqueductal gray (PAG) produces excitation of motoneurons involved in vocalization. It is shown that a specific cell group in the lateral part of the caudal PAG and in the tegmentum just lateral to it projects bilaterally to the nucleus retroambiguus (NRA) in the caudal medulla oblongata. Neurons in the NRA in turn project, via a contralateral pathway through the ventral funiculus of the spinal cord, to the motoneuronal cell groups innervating intercostal and abdominal muscles. In the brainstem, the NRA neurons project to the motoneuronal cell groups innervating mouth-opening and perioral muscles as well as to motoneurons innervating the pharynx, soft palate, and tongue. These results indicate that the projections from PAG via NRA to vocalization motoneurons form the final common pathway in vocalization.

The pathways to lamina I in the medulla oblongata and spinal cord of the cat were traced using horse-radish-peroxidase (HRP) and autoradiographic techniques. The HRP results indicated that several neuronal cell groups in the brain stem and hypothalamus project to the spinal cord throughout its total length. The autoradiographic tracing results demonstrated that the strongest projections to lamina I are derived from the following four areas: the caudal nucleus raphe magnus (NRM), the ventral part of the caudal pontine and NRM, the contralaterally projecting lateral pontine or paralemniscal tegmentum, and the paraventricular nucleus of the hypothalamus. In addition, a limited, especially at lumbosacral levels, distinct projection to lamina I was found to originate in the most caudal part of the medullary tegmentum.

In four rhesus monkeys wheat germ agglutinin-horseradish peroxidase (WGA-HRP) injections were made in the mesencephalic tegmentum. In three cases with injections involving the red nucleus (RN), rubrospinal fibers descended mainly contralaterally to terminate in laminae V, VI and dorsal VII of the spinal cord and in the lateral motoneuronal cell groups at the level of the cervical and lumbosacral enlargements. In all four cases the area of the interstitial nucleus of Cajal (INC) was injected, which resulted in labeled interstitiospinal fibers in the medial part of the ipsilateral ventral funiculus of the spinal cord. The results indicate that there is no major qualitative difference between the mesencephalic (RN and INC) and motor cortical projections to the spinal cord.