Optimization Matters: Guidelines to Improve Representation Learning with Deep Networks

Training deep neural networks is a relevant problem with open questions related to convergence and quality of learned representations. Gradient-based optimization methods are used in practice, but cases of failure and success are still to be investigated. In this context, we set out to better understand the convergence properties of different optimization strategies, under different parameter options. Our results show that (i) feature embeddings are impacted by different optimization settings, (ii) suboptimal results are achieved by the use of default parameters, (iii) significant improvement is obtained by making educated choices of parameters, (iv) learning rate decay should always be considered. Such findings offer guidelines for training and deployment of deep networks.

A Compressive Sensing Approach to Inferring Cognitive Representations with Reverse Correlation

Uncovering cognitive representations is an elusive goal that is increasingly pursued using the reverse correlation method. Employing reverse correlation often entails collecting thousands of stimulus-response pairs from human subjects, a burdensome task that limits the feasibility of many such studies. This methodological barrier can potentially be overcome using recent advances in signal processing designed to improve sampling efficiency, specifically compressive sensing. Here, compressive sensing is shown to be directly compatible with reverse correlation, and a trio of simulations are performed to demonstrate that compressive sensing can improve the accuracy of reconstructed representations while dramatically reducing the required number of samples. This work concludes by outlining the potential of compressive sensing to improve representation reconstruction throughout the field of neuroscience and beyond.

Laying the Groundwork to Investigate Diversity of Life Sciences Reading Lists in Higher Education and Its Link to Awarding Gaps

This study presents a case for decolonising the life sciences curriculum to improve representation of the Black, Asian, and Minority Ethnic (BAME) scholars—a step in eliminating the race “awarding gap”. Here, we investigated diversity among authors in terms of ethnicity and gender of reading lists at the School of Life Sciences, University of Sussex. We show that the reading lists are not diverse and do not represent the demography of the student body. For instance, a disproportionately high number of authors in the reading lists are white 83.40 ± 5.70% (n = 977 authors), male 75.90 ± 5.40% (n = 878 authors), and of European descent. Additionally, our analysis of the geographical locations of publications reveals that a significantly high number of our materials stem from the USA or the UK, whereas the second highest global output of scientific literature (after the USA) comes from China, which is only featured in 1.02% of the reading list. Moreover, we constructively provide potential solutions to decolonise the curriculum of the University of Sussex’s School of Life Sciences by diversifying their reading lists. This study should help to establish a foundation, along with other work that is being conducted, to address the BAME awarding gap and to better showcase the work of women and ethnically underrepresented scientists in history and in modern day.

New observations of NO₂ in the upper troposphere from TROPOMI

Nitrogen oxides (NOx≡NO+NO2) in the NOx-limited upper troposphere (UT) are long-lived and so have a large influence on the oxidizing capacity of the troposphere and formation of the greenhouse gas ozone. Models misrepresent NOx in the UT, and observations to address deficiencies in models are sparse. Here we obtain a year of near-global seasonal mean mixing ratios of NO2 in the UT (450–180 hPa) at 1∘×1∘ by applying cloud-slicing to partial columns of NO2 from TROPOMI. This follows refinement of the cloud-slicing algorithm with synthetic partial columns from the GEOS-Chem chemical transport model. TROPOMI, prior to cloud-slicing, is corrected for a 13 % underestimate in stratospheric NO2 variance and a 50 % overestimate in free-tropospheric NO2 determined by comparison to Pandora total columns at high-altitude free-tropospheric sites at Mauna Loa, Izaña, and Altzomoni and MAX-DOAS and Pandora tropospheric columns at Izaña. Two cloud-sliced seasonal mean UT NO2 products for June 2019 to May 2020 are retrieved from corrected TROPOMI total columns using distinct TROPOMI cloud products that assume clouds are reflective boundaries (FRESCO-S) or water droplet layers (ROCINN-CAL). TROPOMI UT NO2 typically ranges from 20–30 pptv over remote oceans to >80 pptv over locations with intense seasonal lightning. Spatial coverage is mostly in the tropics and subtropics with FRESCO-S and extends to the midlatitudes and polar regions with ROCINN-CAL, due to its greater abundance of optically thick clouds and wider cloud-top altitude range. TROPOMI UT NO2 seasonal means are spatially consistent (R=0.6–0.8) with an existing coarser spatial resolution (5∘ latitude × 8∘ longitude) UT NO2 product from the Ozone Monitoring Instrument (OMI). UT NO2 from TROPOMI is 12–26 pptv more than that from OMI due to increase in NO2 with altitude from the OMI pressure ceiling (280 hPa) to that for TROPOMI (180 hPa), but possibly also due to altitude differences in TROPOMI and OMI cloud products and NO2 retrieval algorithms. The TROPOMI UT NO2 product offers potential to evaluate and improve representation of UT NOx in models and supplement aircraft observations that are sporadic and susceptible to large biases in the UT.

Soil moisture-atmosphere feedbacks mitigate declining water availability in drylands

<p><strong>Global warming alters surface water availability (precipitation minus evapotranspiration, P-E) and hence freshwater resources. However, the influence of land-atmosphere feedbacks on future P-E changes and the underlying mechanisms remain unclear. Here we demonstrate that soil moisture (SM) strongly impacts future P-E changes, especially in drylands, by regulating evapotranspiration and atmospheric moisture inflow. Using modeling and empirical approaches, we find a consistent negative SM feedback on P-E, which may offset ~60% of the decline in dryland P-E otherwise expected in the absence of SM feedbacks. The negative feedback is not caused by atmospheric thermodynamic responses to declining SM, but rather reduced SM, in addition to limiting evapotranspiration, regulates atmospheric circulation and vertical ascent to enhance moisture transport into drylands. This SM effect is a large source of uncertainty in projected dryland P-E changes, underscoring the need to better constrain future SM changes and improve representation of SM-atmosphere processes in models.</strong></p>

Investing in the future: a call for strategies to empower and expand representation of women in neurosurgery worldwide

As progress is gradually being made toward increased representation and retention of women in neurosurgery, the neurosurgical community should elevate effective efforts that may be driving positive change. Here, the authors describe explicit efforts by the neurosurgery community to empower and expand representation of women in neurosurgery, among which they identified four themes: 1) formal mentorship channels; 2) scholarships and awards; 3) training and exposure opportunities; and 4) infrastructural approaches. Ultimately, a data-driven approach is needed to improve representation and empowerment of women in neurosurgery and to best direct the neurosurgical community’s efforts across the globe.

A Liberating Curricula as a Social Responsibility for Promoting Social Justice and Student Success Within the UK Higher Education Institution (HEI)

Integrating corporate social responsibility (CSR) activities as part of a higher education institution (HEI) organisational strategies and practices to address economic and social inequality is no longer a new phenomenon. This promotes increased levels of involvement, choice, and diversity, and is aligned with recent initiatives to widen participation improve representation and promote attainment. CSR may also be encapsulated within frameworks through which HEIs may identify and self-reflect on institutional and cultural barriers that impede minority ethnic (ME) staff and students' progression and attainment. This chapter is informed by discussions concerning CSR within higher education in relation to the aims and objectives of education; student progression and attainment as a university's socially responsible business practice and act of due diligence, to improve representation, progression and success for ME students; curriculum vs. education and the function of a liberating curriculum as a vehicle to enhance academic attainment and promote student success.

Soil moisture-atmosphere feedbacks mitigate declining water availability in drylands

Global warming alters surface water availability (precipitation minus evapotranspiration, P-E) and hence freshwater resources. However, the influence of land-atmosphere feedbacks on future P-E changes and the underlying mechanisms remain unclear. Here we demonstrate that soil moisture (SM) strongly impacts future P-E changes, especially in drylands, by regulating evapotranspiration and atmospheric moisture inflow. Using modeling and empirical approaches, we find a consistent negative SM feedback on P-E, which may offset ~60% of the decline in dryland P-E otherwise expected in the absence of SM feedbacks. The negative feedback is not caused by atmospheric thermodynamic responses to declining SM, but rather reduced SM, in addition to limiting evapotranspiration, regulates atmospheric circulation and vertical ascent to enhance moisture transport into drylands. This SM effect is a large source of uncertainty in projected dryland P-E changes, underscoring the need to better constrain future SM changes and improve representation of SM-atmosphere processes in models.