State-of-the-art visual perception models for a wide range of tasks rely on supervised pretraining. ImageNet classification is the de facto pretraining task for these models. Yet, ImageNet is now nearly ten years old and is by modern standards "small". Even so, relatively little is known about the behavior of pretraining with datasets that are multiple orders of magnitude larger. The reasons are obvious: such datasets are difficult to collect and annotate. In this paper, we present a unique study of transfer learning with large convolutional networks trained to predict hashtags on billions of social media images. Our experiments demonstrate that training for large-scale hashtag prediction leads to excellent results. We show improvements on several image classification and object detection tasks, and report the highest ImageNet-1k single-crop, top-1 accuracy to date: 85.4% (97.6% top-5). We also perform extensive experiments that provide novel empirical data on the relationship between large-scale pretraining and transfer learning performance.

We consider the problem of detecting out-of-distribution images in neural networks. We propose ODIN, a simple and effective method that does not require any change to a pre-trained neural network. Our method is based on the observation that using temperature scaling and adding small perturbations to the input can separate the softmax score distributions between in-and out-of-distribution images, allowing for more effective detection. We show in a series of experiments that ODIN is compatible with diverse network architectures and datasets. It consistently outperforms the baseline approach (Hendrycks & Gimpel, 2017) by a large margin, establishing a new state-of-the-art performance on this task. For example, ODIN reduces the false positive rate from the baseline 34.7% to 4.3% on the DenseNet (applied to CIFAR-10 and Tiny-ImageNet) when the true positive rate is 95%.

No abstract

In this paper, we propose a novel generative model named Stacked Generative Adversarial Networks (SGAN), which is trained to invert the hierarchical representations of a bottom-up discriminative network. Our model consists of a top-down stack of GANs, each learned to generate lower-level representations conditioned on higher-level representations. A representation discriminator is introduced at each feature hierarchy to encourage the representation manifold of the generator to align with that of the bottomup discriminative network, leveraging the powerful discriminative representations to guide the generative model. In addition, we introduce a conditional loss that encourages the use of conditional information from the layer above, and a novel entropy loss that maximizes a variational lower bound on the conditional entropy of generator outputs. We first train each stack independently, and then train the whole model end-to-end. Unlike the original GAN that uses a single noise vector to represent all the variations, our SGAN decomposes variations into multiple levels and gradually resolves uncertainties in the top-down generative process. Based on visual inspection, Inception scores and visual Turing test, we demonstrate that SGAN is able to generate images of much higher quality than GANs without stacking.

Rechargeable solid-state sodium-ion batteries (SSSBs) hold great promise for safer and more energy-dense energy storage. However, the poor electrochemical stability between current sulfide-based solid electrolytes and high-voltage oxide cathodes has limited their long-term cycling performance and practicality. Here, we report the discovery of the ion conductor Na3-xY1-xZrxCl6 (NYZC) that is both electrochemically stable (up to 3.8 V vs. Na/Na+) and chemically compatible with oxide cathodes. Its high ionic conductivity of 6.6 × 10−5 S cm−1 at ambient temperature, several orders of magnitude higher than oxide coatings, is attributed to abundant Na vacancies and cooperative MCl6 rotation, resulting in an extremely low interfacial impedance. A SSSB comprising a NaCrO2 + NYZC composite cathode, Na3PS4 electrolyte, and Na-Sn anode exhibits an exceptional first-cycle Coulombic efficiency of 97.1% at room temperature and can cycle over 1000 cycles with 89.3% capacity retention at 40 °C. These findings highlight the immense potential of halides for SSSB applications.

MicroRNAs are emerging as important regulators of cancer‐related processes. Our studies show that microRNA‐9 (miR‐9) is downregulated in human ovarian cancer relative to normal ovary, and overexpression of miR‐9 suppresses cell growth in vitro. Furthermore, the 3′‐UTR of NF‐κB1 mRNA is found to be regulated directly by miR‐9, demonstrating that NF‐κB1 is a functionally important target of miR‐9 in ovarian cancer cells. When miR‐9 is overexpressed in ovarian cancer cells, the mRNA and protein levels of NF‐κB1 are both suppressed, whereas inhibition of miR‐9 results in an increase in the NF‐κB1 expression level. Ovarian cancer tissues display significantly low expression of miR‐9 and a high level of NF‐κB1 compared with normal tissues, indicating that regulation of NF‐κB1 by miR‐9 is an important mechanism for miR‐9 to inhibit ovarian cancer proliferation.

No abstract

In this work, self-healing polyampholyte hydrogels with high mechanical strength in megapascal order, good resilience, improved toughness, and satisfactory conductivity are fabricated via one-step polymerization of positively charged imidazolium-based ionic liquid monomers containing urea groups and negatively charged 3-sulfopropyl methacrylate potassium salt monomers followed by subsequent dialysis in water. Dialysis can remove partial counter ions in the original hydrogels to strengthen electrostatic interactions between imidazolium and sulfonate groups and improve mechanical strength of the hydrogels. After dialysis for 3 d, the originally soft hydrogels become mechanically robust with a tensile strength of ≈1.3 MPa, strain at break of ≈720%, and toughness of ≈6.7 MJ m −3 . Hydrogen-bonding interactions between urea groups, which act as sacrificial bonds to dissipate energy, are important to improve the mechanical strength and toughness of the hydrogels. More importantly, the hydrogel can automatically heal from physical cut at room temperature with a healing efficiency of ≈91% because of the reversibility of the electrostatic and hydrogen-bonding interactions. Because of the undialyzed salts in the hydrogels, the mechanically robust hydrogels possess a satisfactory ionic conductivity of ≈3 S m −1 at room temperature and can serve as highly flexible and stretchable conductors with self-healing capacity.