TY - GEN AB - The Drosophila Genetic Reference Panel (DGRP) serves as a valuable resource to better understand the genetic landscapes underlying quantitative traits. However, such DGRP studies have so far only focused on nuclear genetic variants. To address this, we sequenced the mitochondrial genomes of >170 DGRP lines, identifying 229 variants including 21 indels and 7 frameshifts. We used our mitochondrial variation data to identify 12 genetically distinct mitochondrial haplotypes, thus revealing important population structure at the mitochondrial level. We further examined whether this population structure was reflected on the nuclear genome by screening for the presence of potential mito-nuclear genetic incompatibilities in the form of significant genotype ratio distortions (GRDs) between mitochondrial and nuclear variants. In total, we detected a remarkable 1,845 mito-nuclear GRDs, with the highest enrichment observed in a 40 kb region around the gene Sex-lethal (Sxl). Intriguingly, downstream phenotypic analyses did not uncover major fitness effects associated with these GRDs, suggesting that a large number of mito-nuclear GRDs may reflect population structure at the mitochondrial level rather than actual genomic incompatibilities. This is further supported by the GRD landscape showing particular large genomic regions associated with a single mitochondrial haplotype. Next, we explored the functional relevance of the detected mitochondrial haplotypes through an association analysis on a set of 259 assembled, non-correlating DGRP phenotypes. We found multiple significant associations with stress- and metabolism-related phenotypes, including food intake in males. We validated the latter observation by reciprocal swapping of mitochondrial genomes from high food intake DGRP lines to low food intake ones. In conclusion, our study uncovered important mitochondrial population structure and haplotype-specific metabolic variation in the DGRP, thus demonstrating the significance of incorporating mitochondrial haplotypes in geno-phenotype relationship studies. AU - Bevers, Roel P.J. AU - Litovchenko, Maria AU - Kapopoulou, Adamandia AU - Braman, Virginie S. AU - Robinson, Matthew Richard AU - Auwerx, Johan AU - Hollis, Brian AU - Deplancke, Bart ID - 7783 T2 - bioRxiv TI - Extensive mitochondrial population structure and haplotype-specific phenotypic variation in the Drosophila Genetic Reference Panel ER - TY - CONF AB - Deep neural networks (DNNs) continue to make significant advances, solving tasks from image classification to translation or reinforcement learning. One aspect of the field receiving considerable attention is efficiently executing deep models in resource-constrained environments, such as mobile or embedded devices. This paper focuses on this problem, and proposes two new compression methods, which jointly leverage weight quantization and distillation of larger teacher networks into smaller student networks. The first method we propose is called quantized distillation and leverages distillation during the training process, by incorporating distillation loss, expressed with respect to the teacher, into the training of a student network whose weights are quantized to a limited set of levels. The second method, differentiable quantization, optimizes the location of quantization points through stochastic gradient descent, to better fit the behavior of the teacher model. We validate both methods through experiments on convolutional and recurrent architectures. We show that quantized shallow students can reach similar accuracy levels to full-precision teacher models, while providing order of magnitude compression, and inference speedup that is linear in the depth reduction. In sum, our results enable DNNs for resource-constrained environments to leverage architecture and accuracy advances developed on more powerful devices. AU - Polino, Antonio AU - Pascanu, Razvan AU - Alistarh, Dan-Adrian ID - 7812 T2 - 6th International Conference on Learning Representations TI - Model compression via distillation and quantization ER - TY - JOUR AB - Feste Alkalicarbonate sind universelle Bestandteile von Passivierungsschichten an Materialien für Interkalationsbatterien, übliche Nebenprodukte in Metall‐O2‐Batterien, und es wird angenommen, dass sie sich reversibel in Metall‐O2 /CO2‐Zellen bilden und zersetzen. In all diesen Kathoden zersetzt sich Li2CO3 zu CO2, sobald es Spannungen >3.8 V vs. Li/Li+ ausgesetzt wird. Beachtenswert ist, dass keine O2‐Entwicklung detektiert wird, wie gemäß der Zersetzungsreaktion 2 Li2CO3 → 4 Li+ + 4 e− + 2 CO2 + O2 zu erwarten wäre. Deswegen war der Verbleib eines der O‐Atome ungeklärt und wurde nicht identifizierten parasitären Reaktionen zugerechnet. Hier zeigen wir, dass hochreaktiver Singulett‐Sauerstoff (1O2) bei der Oxidation von Li2CO3 in einem aprotischen Elektrolyten gebildet und daher nicht als O2 freigesetzt wird. Diese Ergebnisse haben weitreichende Auswirkungen auf die langfristige Zyklisierbarkeit von Batterien: sie untermauern die Wichtigkeit, 1O2 in Metall‐O2‐Batterien zu verhindern, stellen die Möglichkeit einer reversiblen Metall‐O2 /CO2‐Batterie basierend auf einem Carbonat‐Entladeprodukt in Frage und helfen, Grenzflächenreaktivität von Übergangsmetallkathoden mit Li2CO3‐Resten zu erklären. AU - Mahne, Nika AU - Renfrew, Sara E. AU - McCloskey, Bryan D. AU - Freunberger, Stefan Alexander ID - 7983 IS - 19 JF - Angewandte Chemie SN - 0044-8249 TI - Elektrochemische Oxidation von Lithiumcarbonat generiert Singulett-Sauerstoff VL - 130 ER - TY - JOUR AB - The neural code of cortical processing remains uncracked; however, it must necessarily rely on faithful signal propagation between cortical areas. In this issue of Neuron, Joglekar et al. (2018) show that strong inter-areal excitation balanced by local inhibition can enable reliable signal propagation in data-constrained network models of macaque cortex. AU - Stroud, Jake P. AU - Vogels, Tim P ID - 8015 IS - 1 JF - Neuron SN - 0896-6273 TI - Cortical signal propagation: Balance, amplify, transmit VL - 98 ER - TY - JOUR AB - Motor cortex (M1) exhibits a rich repertoire of neuronal activities to support the generation of complex movements. Although recent neuronal-network models capture many qualitative aspects of M1 dynamics, they can generate only a few distinct movements. Additionally, it is unclear how M1 efficiently controls movements over a wide range of shapes and speeds. We demonstrate that modulation of neuronal input–output gains in recurrent neuronal-network models with a fixed architecture can dramatically reorganize neuronal activity and thus downstream muscle outputs. Consistent with the observation of diffuse neuromodulatory projections to M1, a relatively small number of modulatory control units provide sufficient flexibility to adjust high-dimensional network activity using a simple reward-based learning rule. Furthermore, it is possible to assemble novel movements from previously learned primitives, and one can separately change movement speed while preserving movement shape. Our results provide a new perspective on the role of modulatory systems in controlling recurrent cortical activity. AU - Stroud, Jake P. AU - Porter, Mason A. AU - Hennequin, Guillaume AU - Vogels, Tim P ID - 8073 IS - 12 JF - Nature Neuroscience SN - 1097-6256 TI - Motor primitives in space and time via targeted gain modulation in cortical networks VL - 21 ER -