RESUMEN
The N6-methyladenosine (m6A) modification of RNA has been reported to remodel gene expression in response to environmental conditions; however, the biological role of m6A in social insects remains largely unknown. In this study, we explored the role of m6A in the division of labour by worker ants (Solenopsis invicta). We first determined the presence of m6A in RNAs from the brains of worker ants and found that m6A methylation dynamics differed between foragers and nurses. Depletion of m6A methyltransferase or chemical suppression of m6A methylation in foragers resulted in a shift to 'nurse-like' behaviours. Specifically, mRNAs of dopamine receptor 1 (Dop1) and dopamine transporter (DAT) were modified by m6A, and their expression increased dopamine levels to promote the behavioural transition from foragers to nurses. The abundance of Dop1 and DAT mRNAs and their stability were reduced by the inhibition of m6A modification caused by the silencing of Mettl3, suggesting that m6A modification in worker ants modulates dopamine synthesis, which regulates labour division. Collectively, our results provide the first example of the epitranscriptomic regulation of labour division in social insects and implicate m6A regulatory mechanism as a potential novel target for controlling red imported fire ants.
Asunto(s)
Adenosina/análogos & derivados , Hormigas , ARN , Humanos , Animales , Dopamina/genética , Dopamina/metabolismo , Hormigas/genética , ARN Mensajero/metabolismoRESUMEN
Spodoptera frugiperda (J. E. Smith) is an important invasive pest that poses a serious threat to global crop production. Both emamectin benzoate (EB) and diamide insecticides are effective insecticides used to protect against S. frugiperda. Here, 16S rRNA sequencing was used to characterize the gut microbiota in S. frugiperda larvae exposed to EB or tetrachlorantraniliprole (TE). Firmicutes and Proteobacteria were found to be the dominant bacterial phyla present in the intestines of S. frugiperda. Following insecticide treatment, larvae were enriched for species involved in the process of insecticide degradation. High-level alpha and beta diversity indices suggested that exposure to TE and EB significantly altered the composition and diversity of the gastrointestinal microbiota in S. frugiperda. At 24 h post-EB treatment, Burkholderia-Caballeronia-Paraburkholderia abundance was significantly increased relative to the control group, with significant increases in Stenotrophobacter, Nitrospira, Blastocatella, Sulfurifustis, and Flavobacterium also being evident in these larvae. These microbes may play a role in the degradation or detoxification of EB and TE, although further work will be needed to explore the mechanisms underlying such activity. Overall, these findings will serve as a theoretical foundation for subsequent studies of the relationship between the gut microbiota and insecticide resistance in S. frugiperda (J. E. Smith) (Lepidoptera: Noctuidae).
Asunto(s)
Microbioma Gastrointestinal , Insecticidas , Animales , Spodoptera/genética , Insecticidas/farmacología , ARN Ribosómico 16S/genética , Larva , Resistencia a los Insecticidas/genéticaRESUMEN
The insulin signalling pathway plays a crucial role in regulating the metabolism of sugars, fats and proteins in cells, thereby affecting the growth, metabolism, reproduction and ageing of organisms. However, little is known about the functions of long non-coding RNAs (lncRNAs) in the regulation of insulin receptors under stress conditions in insects. In this study, we showed that insulin receptor-associated lncRNA (IRAR) regulates insulin receptor transcripts in response to nutritional stress in Drosophila melanogaster. Genome editing by CRISPR-Cas9 showed reduced sensitivity of IRAR mutants to environmental nutritional changes. In contrast, the sensitivity of mutants overexpressing tubulin-gal4 > IRAR increased under low nutrition. The pupation and eclosion timings in IRAR mutants were significantly delayed with an increase in insulin concentration compared with that in the w1118 group. In addition, the expression pattern of IRAR was almost consistent with that of the four transcripts of the insulin receptor from the embryonic period to the adult period. RNA immunoprecipitation assay showed the direct regulation of insulin receptor transcripts by IRAR to the through FOXO binding under nutritional stress. To our knowledge, this is the first study that describes a model of lncRNA-mediated development regulation through insulin receptor transcripts.
Asunto(s)
Proteínas de Drosophila , ARN Largo no Codificante , Animales , Proteínas de Drosophila/genética , Proteínas de Drosophila/metabolismo , Drosophila melanogaster/fisiología , Insulina/genética , Insulina/metabolismo , ARN Largo no Codificante/genética , ARN Largo no Codificante/metabolismo , Receptor de Insulina/genética , Receptor de Insulina/metabolismoRESUMEN
The synthesis of previously unknown perovskite (CH3 NH3 )2 PdCl4 is reported. Despite using an organic cation with the smallest possible alkyl group, a 2D organic-inorganic layered Pd-based perovskites was still formed. This demonstrates that Pd-based 2D perovskites can be obtained even if the size of the organic cation is below the size limit predicted by the Goldschmidt tolerance-factor formula. The (CH3 NH3 )2 PdCl4 phase has a bulk resistivity of 1.4â Ω cm, a direct optical gap of 2.22â eV, and an absorption coefficient on the order of 104 â cm-1 . XRD measurements suggest that the compound is moderately stable in air, an important advantage over several existing organic-inorganic perovskites that are prone to phase degradation problems when exposed to the atmosphere. Given the recent interest in organic-inorganic perovskites, the synthesis of this new Pd-based organic-inorganic perovskite may be helpful in the preparation and understanding of other organic-inorganic perovskites.
RESUMEN
Inferring 3D human motion is fundamental in many applications, including understanding human activity and analyzing one's intention. While many fruitful efforts have been made to human motion prediction, most approaches focus on pose-driven prediction and inferring human motion in isolation from the contextual environment, thus leaving the body location movement in the scene behind. However, real-world human movements are goal-directed and highly influenced by the spatial layout of their surrounding scenes. In this paper, instead of planning future human motion in a "dark" room, we propose a Multi-Condition Latent Diffusion network (MCLD) that reformulates the human motion prediction task as a multi-condition joint inference problem based on the given historical 3D body motion and the current 3D scene contexts. Specifically, instead of directly modeling joint distribution over the raw motion sequences, MCLD performs a conditional diffusion process within the latent embedding space, characterizing the cross-modal mapping from the past body movement and current scene context condition embeddings to the future human motion embedding. Extensive experiments on large-scale human motion prediction datasets demonstrate that our MCLD achieves significant improvements over the state-of-the-art methods on both realistic and diverse predictions.
Asunto(s)
Movimiento , Humanos , Movimiento/fisiología , Algoritmos , Redes Neurales de la Computación , Grabación en Video/métodos , Imagenología Tridimensional/métodos , Procesamiento de Imagen Asistido por Computador/métodosRESUMEN
The insect gustatory system participates in identifying potential food sources and avoiding toxic compounds. During this process, gustatory receptors (GRs) recognize feeding stimulant and deterrent compounds. However, the GRs involved in recognizing stimulant and deterrent compounds in the red imported fire ant, Solenopsis invicta, remain unknown. Therefore, we conducted a study on the genes SinvGR1, SinvGR32b, and SinvGR28a to investigate the roles of GRs in detecting feeding stimulant and deterrent compounds. In this current study, we found that sucrose and fructose are feeding stimulants and the bitter compound quinine is a feeding deterrent. The fire ant workers showed significant behavior changes to avoid the bitter taste in feeding stimulant compounds. Reverse transcription quantitative real-time polymerase chain reaction results from developmental stages showed that the SinvGR1, SinvGR32b, and SinvGR28a genes were highly expressed in fire ant workers. Tissue-specific expression profiles indicated that SinvGR1, SinvGR32b, and SinvGR28a were specifically expressed in the antennae and foreleg tarsi of workers, whereas SinvGR32b gene transcripts were also highly accumulated in the male antennae. Furthermore, the silencing of SinvGR1 or SinvGR32b alone and the co-silencing of both genes disrupted worker stimulation and feeding on sucrose and fructose. The results also showed that SinvGR28a is required for avoiding quinine, as workers with knockdown of the SinvGR28a gene failed to avoid and fed on quinine. This study first identified stimulant and deterrent compounds of fire ant workers and then the GRs involved in the taste recognition of these compounds. This study could provide potential target gustatory genes for the control of the fire ant.
Asunto(s)
Hormigas , Gusto , Masculino , Animales , Hormigas de Fuego , Quinina/farmacología , Quinina/metabolismo , Hormigas/fisiología , Receptores de Superficie Celular/genética , Receptores de Superficie Celular/metabolismo , Fructosa/metabolismo , Sacarosa/metabolismoRESUMEN
Juvenile hormone (JH) acts in the regulation of caste differentiation between queens and workers (i.e., with or without reproductive capacity) during vitellin synthesis and oogenesis in social insects. However, the regulatory mechanisms have not yet been elucidated. Here, we identified a highly expressed microRNA (miRNA), miR-1175-3p, in the red imported fire ant, Solenopsis invicta. We found that miR-1175-3p is prominently present in the fat bodies and ovaries of workers. Furthermore, miR-1175-3p interacts with its target gene, broad-complex core (Br-C), in the fat bodies. By utilizing miR-1175-3p agomir, we successfully suppressed the expression of the Br-C protein in queens, resulting in reduced vitellogenin expression, fewer eggs, and poorly developed ovaries. Conversely, decreasing miR-1175-3p levels led to the increased expression of Br-C and vitellogenin in workers, triggering the "re-development" of the ovaries. Moreover, when queens were fed with JH, the expression of miR-1175-3p decreased, whereas the expression of vitellogenin-2 and vitellogenin-3 increased. Notably, the suppression of fertility in queens caused by treatment with agomir miR-1175-3p was completely rescued by the increased vitellogenin expression induced by being fed with JH. These results suggest the critical role of miR-1175-3p in JH-regulated reproduction, shedding light on the molecular mechanism underlying miRNA-mediated fecundity in social insects and providing a novel strategy for managing S. invicta.
Asunto(s)
Hormigas , MicroARNs , Animales , Vitelogeninas/genética , Vitelogeninas/metabolismo , Hormigas de Fuego , Hormonas Juveniles/metabolismo , Hormigas/fisiología , Reproducción , MicroARNs/genética , MicroARNs/metabolismoRESUMEN
Hot water mound drench treatment has broad application prospects in the control of the red imported fire ant (RIFA), Solenopsis invicta Buren (Hymenoptera: Formicidae). However, much work still needs to be carried out to provide a theoretical basis and technical support for the use of this method against RIFAs under field conditions. In this study, we monitored the temperature changes at different depths within RIFA nests during laboratory-simulated hot water mound drench experiments and evaluated the lethal effect of hot water treatment on RIFAs. Furthermore, the targeted removal effect of hot water mound drench treatment on RIFA nests under field conditions was evaluated. Results indicated that the temperature at depths of 5, 15, and 25 cm inside the simulated ant nests was higher than 51.1 °C within 30 min after treatment, resulting in a 100% mortality rate for RIFAs at all tested depths. Under field conditions, when nests were disturbed, the percentage of RIFAs crawling out of their nests gradually increased with time after disturbance, reached its maximum value at 25 or 30 s after the disturbance, and then gradually decreased. Single hot water mound drench treatment (each ant nest was treated with 17.8-21.6 liter of hot water at a temperature of 97-100 °C) can significantly reduce the RIFA population in ant nests and lead to a 72.7% reduction in the number of surviving ant nests. However, the safety, operability, and timelines of hot water mound drench treatment for RIFA field control still need further investigation.
Asunto(s)
Hormigas de Fuego , Calor , Control de Insectos , Animales , Control de Insectos/métodos , AguaRESUMEN
Representation learning has significantly been developed with the advance of contrastive learning methods. Most of those methods are benefited from various data augmentations that are carefully designated to maintain their identities so that the images transformed from the same instance can still be retrieved. However, those carefully designed transformations limited us to further explore the novel patterns exposed by other transformations. Meanwhile, as shown in our experiments, direct contrastive learning for stronger augmented images can not learn representations effectively. Thus, we propose a general framework called Contrastive Learning with Stronger Augmentations (CLSA) to complement current contrastive learning approaches. Here, the distribution divergence between the weakly and strongly augmented images over the representation bank is adopted to supervise the retrieval of strongly augmented queries from a pool of instances. Experiments on the ImageNet dataset and downstream datasets showed the information from the strongly augmented images can significantly boost the performance. For example, CLSA achieves top-1 accuracy of 76.2% on ImageNet with a standard ResNet-50 architecture with a single-layer classifier fine-tuned, which is almost the same level as 76.5% of supervised results.
RESUMEN
Semi-supervised dense prediction tasks, such as semantic segmentation, can be greatly improved through the use of contrastive learning. However, this approach presents two key challenges: selecting informative negative samples from a highly redundant pool and implementing effective data augmentation. To address these challenges, we present an adversarial contrastive learning method specifically for semi-supervised semantic segmentation. Direct learning of adversarial negatives is adopted to retain discriminative information from the past, leading to higher learning efficiency. Our approach also leverages an advanced data augmentation strategy called AdverseMix, which combines information from under-performing classes to generate more diverse and challenging samples. Additionally, we use auxiliary labels and classifiers to prevent over-adversarial negatives from affecting the learning process. Our experiments on the Pascal VOC and Cityscapes datasets demonstrate that our method outperforms the state-of-the-art by a significant margin, even when using a small fraction of labeled data.
RESUMEN
As a representative self-supervised method, contrastive learning has achieved great successes in unsupervised training of representations. It trains an encoder by distinguishing positive samples from negative ones given query anchors. These positive and negative samples play critical roles in defining the objective to learn the discriminative encoder, avoiding it from learning trivial features. While existing methods heuristically choose these samples, we present a principled method where both positive and negative samples are directly learnable end-to-end with the encoder. We show that the positive and negative samples can be cooperatively and adversarially learned by minimizing and maximizing the contrastive loss, respectively. This yields cooperative positives and adversarial negatives with respect to the encoder, which are updated to continuously track the learned representation of the query anchors over mini-batches. The proposed method achieves 71.3% and 75.3% in top-1 accuracy respectively over 200 and 800 epochs of pre-training ResNet-50 backbone on ImageNet1K without tricks such as multi-crop or stronger augmentations. With Multi-Crop, it can be further boosted into 75.7%. The source code and pre-trained model are released in https://github.com/maple-research-lab/caco.
RESUMEN
Human parsing aims to segment each pixel of the human image with fine-grained semantic categories. However, current human parsers trained with clean data are easily confused by numerous image corruptions such as blur and noise. To improve the robustness of human parsers, in this paper, we construct three corruption robustness benchmarks, termed LIP-C, ATR-C, and Pascal-Person-Part-C, to assist us in evaluating the risk tolerance of human parsing models. Inspired by the data augmentation strategy, we propose a novel heterogeneous augmentation-enhanced mechanism to bolster robustness under commonly corrupted conditions. Specifically, two types of data augmentations from different views, i.e., image-aware augmentation and model-aware image-to-image transformation, are integrated in a sequential manner for adapting to unforeseen image corruptions. The image-aware augmentation can enrich the high diversity of training images with the help of common image operations. The model-aware augmentation strategy that improves the diversity of input data by considering the model's randomness. The proposed method is model-agnostic, and it can plug and play into arbitrary state-of-the-art human parsing frameworks. The experimental results show that the proposed method demonstrates good universality which can improve the robustness of the human parsing models and even the semantic segmentation models when facing various image common corruptions. Meanwhile, it can still obtain approximate performance on clean data.
RESUMEN
Deep learning-based models have been shown to outperform human beings in many computer vision tasks with massive available labeled training data in learning. However, humans have an amazing ability to easily recognize images of novel categories by browsing only a few examples of these categories. In this case, few-shot learning comes into being to make machines learn from extremely limited labeled examples. One possible reason why human beings can well learn novel concepts quickly and efficiently is that they have sufficient visual and semantic prior knowledge. Toward this end, this work proposes a novel knowledge-guided semantic transfer network (KSTNet) for few-shot image recognition from a supplementary perspective by introducing auxiliary prior knowledge. The proposed network jointly incorporates vision inferring, knowledge transferring, and classifier learning into one unified framework for optimal compatibility. A category-guided visual learning module is developed in which a visual classifier is learned based on the feature extractor along with the cosine similarity and contrastive loss optimization. To fully explore prior knowledge of category correlations, a knowledge transfer network is then developed to propagate knowledge information among all categories to learn the semantic-visual mapping, thus inferring a knowledge-based classifier for novel categories from base categories. Finally, we design an adaptive fusion scheme to infer the desired classifiers by effectively integrating the above knowledge and visual information. Extensive experiments are conducted on two widely used Mini-ImageNet and Tiered-ImageNet benchmarks to validate the effectiveness of KSTNet. Compared with the state of the art, the results show that the proposed method achieves favorable performance with minimal bells and whistles, especially in the case of one-shot learning.
RESUMEN
BACKGROUND: In social insects, the labor division of workers is ubiquitous and controlled by genetic and environmental factors. However, how they modulate this coordinately remains poorly understood. RESULTS: We report miR-279c-5p participation in insulin synthesis and behavioral transition by negatively regulating Rab8A in Solenopsis invicta. Eusocial specific miR-279c-5p is age-associated and highly expressed in nurse workers, and localized in the cytoplasm of neurons, where it is partly co-localized with its target, Rab8A. We determined that miR-279c-5p agomir suppressed Rab8A expression in forager workers, consequently decreasing insulin content, resulting in the behavioral shift to 'nurse-like' behaviors, while the decrease in miR-279c-5p increased Rab8A expression and increased insulin content in nurse workers, leading to the behavioral shift to 'foraging-like' behaviors. Moreover, insulin could rescue the 'foraging behavior' induced by feeding miR-279c-5p to nurse workers. The overexpression and suppression of miR-279c-5p in vivo caused an obvious behavioral transition between foragers and nurses, and insulin synthesis was affected by miR-279c-5p by regulating the direct target Rab8A. CONCLUSION: We first report that miR-279c-5p is a novel regulator that promotes labor division by negatively regulating the target gene Rab8A by controlling insulin production in ants. This miRNA-mediated mechanism is significant for understanding the behavioral plasticity of social insects between complex factors and potentially provides new targets for controlling red imported fire ants. © 2023 Society of Chemical Industry.
Asunto(s)
Hormigas , MicroARNs , Humanos , Animales , Hormigas/fisiología , Insulina/metabolismo , Conducta Alimentaria , MicroARNs/genética , MicroARNs/metabolismoRESUMEN
The brown planthopper (BPH) is the most serious pest causing yield losses in rice. MicroRNAs (miRNAs) are emerging as key modulators of plant-pest interactions. In the study, we found that osa-miR162a is induced in response to BPH attack in the seedling stage and tunes rice resistance to the BPH via the α-linolenic acid metabolism pathway as indicated by gas chromatography/liquid chromatography-mass spectrometry analysis. Overexpression of osa-miR162a inhibited the development and growth of the BPH and simultaneously reduced the release of 3-hexenal and 3-hexen-1-ol to block host recognition in the BPH. Moreover, knockdown of OsDCL1, which is targeted by osa-miR162a, inhibited α-linolenic acid metabolism to enhance the resistance to the BPH, which was similar to that in miR162a-overexpressing plants. Our study revealed a novel defense mechanism mediated by plant miRNAs developed during the long-term evolution of plant-host interaction, provided new ideas for the identification of rice resistance resources, and promoted a better understanding of pest control.
Asunto(s)
Hemípteros , MicroARNs , Oryza , Ácido alfa-Linolénico , Regulación de la Expresión Génica de las Plantas , Hemípteros/fisiología , MicroARNs/genética , MicroARNs/metabolismo , Oryza/química , AnimalesRESUMEN
Recently, deep convolution neural networks (CNNs) steered face super-resolution methods have achieved great progress in restoring degraded facial details by joint training with facial priors. However, these methods have some obvious limitations. On the one hand, multi-task joint learning requires additional marking on the dataset, and the introduced prior network will significantly increase the computational cost of the model. On the other hand, the limited receptive field of CNN will reduce the fidelity and naturalness of the reconstructed facial images, resulting in suboptimal reconstructed images. In this work, we propose an efficient CNN-Transformer Cooperation Network (CTCNet) for face super-resolution tasks, which uses the multi-scale connected encoder-decoder architecture as the backbone. Specifically, we first devise a novel Local-Global Feature Cooperation Module (LGCM), which is composed of a Facial Structure Attention Unit (FSAU) and a Transformer block, to promote the consistency of local facial detail and global facial structure restoration simultaneously. Then, we design an efficient Feature Refinement Module (FRM) to enhance the encoded features. Finally, to further improve the restoration of fine facial details, we present a Multi-scale Feature Fusion Unit (MFFU) to adaptively fuse the features from different stages in the encoder procedure. Extensive evaluations on various datasets have assessed that the proposed CTCNet can outperform other state-of-the-art methods significantly. Source code will be available at https://github.com/IVIPLab/CTCNet.
Asunto(s)
Aprendizaje , Redes Neurales de la Computación , Programas Informáticos , Procesamiento de Imagen Asistido por ComputadorRESUMEN
Introduction: In recent decades, single-cell sequencing technology has developed rapidly and used widely in various fields of life sciences, especially for the detection of immune cells. A bibliometric analysis of single-cell sequencing research work on immune cells published during the 2011-2021 period should provide new insight on the use of single-cell sequencing. Methods: We screened 1,460 publications on single-cell sequencing on immune cells according to the publication date, article type, language, and country. Reults: The United States published the first and largest number of articles, while China's research started relatively late, but ranked second in the number of publications. T cells were the most commonly studied immune cells by single-cell sequencing, followed by mononuclear macrophages. Cancer biology was the most common field of immune cell research by single-cell sequencing. Single-cell sequencing studies using γδ T cells were mainly in the fields of cancer biology and cell development, and focused over time from cell surface receptor to cell function. Through in-depth analysis of the articles on single-cell sequencing of T cells in the oncology field, our analysis found that immunotherapy and tumor microenvironment were the most popular research directions in recent years. Discussion: The combination of DNA damage repair and immunotherapy seems to provide a new strategy for cancer therapy.
RESUMEN
Representation learning with small labeled data have emerged in many problems, since the success of deep neural networks often relies on the availability of a huge amount of labeled data that is expensive to collect. To address it, many efforts have been made on training sophisticated models with few labeled data in an unsupervised and semi-supervised fashion. In this paper, we will review the recent progresses on these two major categories of methods. A wide spectrum of models will be categorized in a big picture, where we will show how they interplay with each other to motivate explorations of new ideas. We will review the principles of learning the transformation equivariant, disentangled, self-supervised and semi-supervised representations, all of which underpin the foundation of recent progresses. Many implementations of unsupervised and semi-supervised generative models have been developed on the basis of these criteria, greatly expanding the territory of existing autoencoders, generative adversarial nets (GANs) and other deep networks by exploring the distribution of unlabeled data for more powerful representations. We will discuss emerging topics by revealing the intrinsic connections between unsupervised and semi-supervised learning, and propose in future directions to bridge the algorithmic and theoretical gap between transformation equivariance for unsupervised learning and supervised invariance for supervised learning, and unify unsupervised pretraining and supervised finetuning. We will also provide a broader outlook of future directions to unify transformation and instance equivariances for representation learning, connect unsupervised and semi-supervised augmentations, and explore the role of the self-supervised regularization for many learning problems.
Asunto(s)
Algoritmos , Macrodatos , Redes Neurales de la Computación , Aprendizaje Automático SupervisadoRESUMEN
In this paper, we propose the K-Shot Contrastive Learning (KSCL) of visual features by applying multiple augmentations to investigate the sample variations within individual instances. It aims to combine the advantages of inter-instance discrimination by learning discriminative features to distinguish between different instances, as well as intra-instance variations by matching queries against the variants of augmented samples over instances. Particularly, for each instance, it constructs an instance subspace to model the configuration of how the significant factors of variations in K-shot augmentations can be combined to form the variants of augmentations. Given a query, the most relevant variant of instances is then retrieved by projecting the query onto their subspaces to predict the positive instance class. This generalizes the existing contrastive learning that can be viewed as a special one-shot case. An eigenvalue decomposition is performed to configure instance subspaces, and the embedding network can be trained end-to-end through the differentiable subspace configuration. Experiment results demonstrate the proposed K-shot contrastive learning achieves superior performances to the state-of-the-art unsupervised methods.
Asunto(s)
Algoritmos , AprendizajeRESUMEN
Transformation equivariant representations (TERs) aim to capture the intrinsic visual structures that equivary to various transformations by expanding the notion of translation equivariance underlying the success of convolutional neural networks (CNNs). For this purpose, we present both deterministic AutoEncoding Transformations (AET) and probabilistic AutoEncoding Variational Transformations (AVT) models to learn visual representations from generic groups of transformations. While the AET is trained by directly decoding the transformations from the learned representations, the AVT is trained by maximizing the joint mutual information between the learned representation and transformations. This results in generalized TERs (GTERs) equivariant against transformations in a more general fashion by capturing complex patterns of visual structures beyond the conventional linear equivariance under a transformation group. The presented approach can be extended to (semi-)supervised models by jointly maximizing the mutual information of the learned representation with both labels and transformations. Experiments demonstrate the proposed models outperform the state-of-the-art models in both unsupervised and (semi-)supervised tasks. Moreover, we show that the unsupervised representation can even surpass the fully supervised representation pretrained on ImageNet when they are fine-tuned for the object detection task.