SoxB family genes delay regeneration and cause abnormal movement in Dugesia japonica.

SoxB subfamily is an important branch of Sox family and plays a key role in animal physiological process, but little is known about their function in planarian regeneration. This study aims to evaluate the function of DjSoxB family genes in intact and regenerating planarians Dugesia japonica. Here, we amplify the full-length cDNA of DjSoxB1 and DjSoxB2 in D. japonica by rapid amplification of the cDNA ends (RACE), detect the expression of DjSoxB family genes in planarian. The results show that DjSoxBs are expressed in parenchymal tissue and the hybridization signals partially disappear after irradiation indicates DjSoxB family genes are expressed in neoblasts. After the RNA interference (RNAi) of DjSoxB1, DjSoxB2 and DjSoxB3 separately, the numbers of proliferative cells are all reduced that causes planarians show slower growth of blastema in the early stage of regeneration, and nerves of planarians are affected that the movement speed of planarians decreases in varying degrees. Specially, planarians in the DjSoxB3 RNAi group show shrinkage and twisting. Overall, this study reveals that DjSoxB family genes play a role in cell proliferation during regeneration. They also play an important role in the maintenance of normal nerve function and nerve regeneration. These results provide directions for the functional study of SoxB family genes and provide an important foundation for planarian regeneration.

Progressive enhanced photodynamic therapy and enhanced chemotherapy fighting against malignant tumors with sequential drug release.

During the process of malignant tumor treatment, photodynamic therapy (PDT) exerts poor efficacy due to the hypoxic environment of the tumor cells, and long-time chemotherapy reduces the sensitivity of tumor cells to chemotherapy drugs due to the presence of drug-resistant proteins on the cell membranes for drug outward transportation. Therefore, we reported a nano platform based on mesoporous silica coated with polydopamine (MSN@PDA) loading PDT enhancer MnO2, photosensitizer indocyanine green (ICG) and chemotherapeutic drug doxorubicin (DOX) (designated as DMPIM) to achieve a sequential release of different drugs to enhance treatment of malignant tumors. MSN was first synthesized by a template method, then DOX was loaded into the mesoporous channels of MSN, and locked by the PDA coating. Next, ICG was modified by π-π stacking on PDA, and finally, MnO2layer was accumulated on the surface of DOX@MSN@PDA- ICG@MnO2, achieving orthogonal loading and sequential release of different drugs. DMPIM first generated oxygen (O2) through the reaction between MnO2and H2O2after entering tumor cells, alleviating the hypoxic environment of tumors and enhancing the PDT effect of sequentially released ICG. Afterwards, ICG reacted with O2in tumor tissue to produce reactive oxygen species, promoting lysosomal escape of drugs and inactivation of p-glycoprotein (p-gp) on tumor cell membranes. DOX loaded in the MSN channels exhibited a delay of approximately 8 h after ICG release to exert the enhanced chemotherapy effect. The drug delivery system achieved effective sequential release and multimodal combination therapy, which achieved ideal therapeutic effects on malignant tumors. This work offers a route to a sequential drug release for advancing the treatment of malignant tumors.

UPLC-Triple-TOF-MS-based serum metabonomic revealed the alleviating effect of QingYan Formula on perimenopausal syndrome rats.

The study aimed to investigate the relieving effect of QingYan Formula (QYF) in treating perimenopausal syndrome. A combination of metabonomic analysis and in vitro pharmacodynamic experiments was employed to achieve this objective.Over a period of 12 weeks, ovariectomized (OVX) rats were orally administered QYF's 70 % ethanol extract or estradiol valerate (EV). The results demonstrate that QYF restored the estrous cycle of ovariectomized rats and exhibited significant estrogenic activity, as indicated by reversal of uterine and vagina atrophy, improvement of serum estradiol level and decrease of serum luteinizing hormone(LH) level. Additionally, QYF administration effectively reduced high bone turnover and repaired trabecular microstructure damage. Metabonomic analysis of the OVX rats treated with QYF revealed the identification of 55 different metabolites in the serum, out of which 35 may be potential biomarkers. QYF could regulate the disturbed metabolic pathways including the Biosynthesis of unsaturated fatty acids, arachidonic acid metabolism, bile secretion, and steroid hormone biosynthesis. PI3KCA, SRC, and MAPK3 are potential therapeutic targets for QYF therapeutic effects. These findings support the efficacy of QYF in alleviating perimenopausal syndrome and regulating lipid metabolic disorders in OVX rats.

Cenozoic Indo-Pacific warm pool controlled by both atmospheric CO2 and paleogeography.

The Indo-Pacific warm pool (IPWP) is crucial for regional and global climates. However, the development of the IPWP and its effect on the regional climate during the Cenozoic remain unclear. Here, using a compilation of sea surface temperature (SST) records (mainly since the middle Miocene) and multimodel paleoclimate simulations, our results indicated that the extent, intensity and warmest temperature position of the IPWP changed markedly during the Cenozoic. Specifically, its extent decreased, its intensity weakened, and its warmest temperature position shifted from the Indian to western Pacific Ocean over time. The atmospheric CO2 dominated its extent and intensity, while paleogeography, by restricting the distribution of the Indian Ocean and the width of the tropical seaways, controlled the shift in its warmest temperature position. In particular, the eastward shift to the western Pacific Ocean from the middle to late Miocene inferred from compiled SST records likely resulted from the constriction of tropical seaways. Furthermore, by changing the atmospheric thermal structure and atmospheric circulation, the reduced extent and intensity of the IPWP decreased the annual precipitation in the western Indian Ocean, eastern Asia and Australia, while the shift in the warmest temperature position from the Indian to western Pacific Ocean promoted aridification in Australia. Qualitative model-data agreements are obtained for both the IPWP SST and regional climate. From the perspective of past warm climates with high concentrations of atmospheric CO2, the expansion and strengthening of the IPWP will occur in a warmer future and favor excessive precipitation in eastern Asia and Australia.

Towards Open Vocabulary Learning: A Survey.

In the field of visual scene understanding, deep neural networks have made impressive advancements in various core tasks like segmentation, tracking, and detection. However, most approaches operate on the close-set assumption, meaning that the model can only identify pre-defined categories that are present in the training set. Recently, open vocabulary settings were proposed due to the rapid progress of vision language pre-training. These new approaches seek to locate and recognize categories beyond the annotated label space. The open vocabulary approach is more general, practical, and effective than weakly supervised and zero-shot settings. This paper thoroughly reviews open vocabulary learning, summarizing and analyzing recent developments in the field. In particular, we begin by juxtaposing open vocabulary learning with analogous concepts such as zero-shot learning, open-set recognition, and out-of-distribution detection. Subsequently, we examine several pertinent tasks within the realms of segmentation and detection, encompassing long-tail problems, few-shot, and zero-shot settings. As a foundation for our method survey, we first elucidate the fundamental principles of detection and segmentation in close-set scenarios. Next, we examine various contexts where open vocabulary learning is employed, pinpointing recurring design elements and central themes. This is followed by a comparative analysis of recent detection and segmentation methodologies in commonly used datasets and benchmarks. Our review culminates with a synthesis of insights, challenges, and discourse on prospective research trajectories. To our knowledge, this constitutes the inaugural exhaustive literature review on open vocabulary learning.

Nanoparticles loaded withß-Lapachone and Fe3+exhibit enhanced chemodynamic therapy by producing H2O2through cascaded amplification.

The rapid, irreversible change of active Fe2+to inactive Fe3+after the Fenton reaction occurring reduces the chemodynamic therapeutic (CDT) effect. Therefore, manipulation of the tumor microenvironment to provide sufficient hydrogen peroxide (H2O2) while maintaining metal ion catalyst activity is critical for effective CDT. Here,ß-Lapachone (LPC) was loaded by mesoporous silica nanoparticles (MSNs) and coated with polydopamine (PDA) to further chelate Fe3+and link aptamer AS1411, and a pH-controlled released, chemotherapy-photothermal therapy (PTT)-enhanced CDT-small molecule therapy combination drug delivery system with passive and active tumor targeting was engineered (designated asß-LPC@MSN@PDA/Fe3+-AS1411, LMPFA). The results showed that LFMPA nanoparticles massively accumulated in tumor tissues to achieve tumor targeting through AS1411 mediating and enhanced permeability and retention (EPR) effect. Subsequently, PDA released Fe3+and LPC through acid response to exhibited CDT and chemotherapeutic therapy. Meanwhile, the photothermal effect of PDA promoted the release of LPC from the pores of MSN. LPC exerted chemotherapy effect and cyclically producing of H2O2by the catalysis of NQO1, which enhanced the CDT activated by Fe3+. In addition, while serving as a targeted ligand, AS1411 could also exhibit a small molecule therapeutic effect by binding to nucleoli of tumor cells. This unique nano delivery system achieved the combination of chemotherapy, PTT, enhanced CDT and small molecule therapy, and fought against malignant tumors synergistically through multi-target and multi-dimension.

Multilayer nanodrug delivery system with spatiotemporal drug release improves tumor microenvironment for synergistic anticancer therapy.

The inflammatory response is one of the general symptoms that accompany tumorigenesis, the pro-inflammatory factors cyclooxygenase-2 (COX-2) and COX-2-derived prostaglandin-2 (PGE-2) in the inflammatory environment surrounding tumors possess promoting tumor development, metastasis and angiogenesis effects. In addition, the hypoxic environment of tumors severely limits the effectiveness of photodynamic therapy (PDT). In this study, a universal extracellular-intracellular 'on-demand' release nanomedicine DOX@PDA-ICG@MnO2@GN-CEL was developed for the combined fight against malignant tumors using a spatiotemporal controlled gelatin coated polydopamine (PDA@GN) as the carrier and loaded with the chemotherapeutic drug doxorubicin (DOX), the photosensitizer indocyanine green (ICG), the PDT enhancer MnO2and the anti-inflammatory drug celecoxib (CEL) individually. Our results showed that DOX@PDA-ICG@MnO2@GN-CEL could release CEL extracellularly by matrix metalloproteinase-2 response and inhibit the COX-2/PGE-2 pathway, reduce chemotherapy resistance and attenuate the concurrent inflammation. After entering the tumor cells, the remaining DOX@PDA-ICG@MnO2released DOX, ICG and MnO2intracellularly through PDA acid response. MnO2promoted the degradation of endogenous H2O2to generate oxygen under acidic conditions to alleviate the tumor hypoxic environment, enhance PDT triggered by ICG. PDA and ICG exhibited photothermal therapy synergistically, and DOX exerted chemotherapy with reduced chemotherapy resistance. The dual responsive drug release switch enabled the chemotherapeutic, photothermal, photodynamic and anti-inflammatory drugs precisely acted on different sites of tumor tissues and realized a promising multimodal combination therapy.

Direct probing of single-molecule chemiluminescent reaction dynamics under catalytic conditions in solution.

Chemical reaction kinetics can be evaluated by probing dynamic changes of chemical substrates or physical phenomena accompanied during the reaction process. Chemiluminescence, a light emitting exoenergetic process, involves random reaction positions and kinetics in solution that are typically characterized by ensemble measurements with nonnegligible average effects. Chemiluminescent reaction dynamics at the single-molecule level remains elusive. Here we report direct imaging of single-molecule chemiluminescent reactions in solution and probing of their reaction dynamics under catalytic conditions. Double-substrate Michaelis-Menten type of catalytic kinetics is found to govern the single-molecule reaction dynamics in solution, and a heterogeneity is found among different catalyst particles and different catalytic sites on a single particle. We further show that single-molecule chemiluminescence imaging can be used to evaluate the thermodynamics of the catalytic system, resolving activation energy at the single-particle level. Our work provides fundamental insights into chemiluminescent reactions and offers an efficient approach for evaluating catalysts.

Problem-Dependent Power of Quantum Neural Networks on Multiclass Classification.

Quantum neural networks (QNNs) have become an important tool for understanding the physical world, but their advantages and limitations are not fully understood. Some QNNs with specific encoding methods can be efficiently simulated by classical surrogates, while others with quantum memory may perform better than classical classifiers. Here we systematically investigate the problem-dependent power of quantum neural classifiers (QCs) on multiclass classification tasks. Through the analysis of expected risk, a measure that weighs the training loss and the generalization error of a classifier jointly, we identify two key findings: first, the training loss dominates the power rather than the generalization ability; second, QCs undergo a U-shaped risk curve, in contrast to the double-descent risk curve of deep neural classifiers. We also reveal the intrinsic connection between optimal QCs and the Helstrom bound and the equiangular tight frame. Using these findings, we propose a method that exploits loss dynamics of QCs to estimate the optimal hyperparameter settings yielding the minimal risk. Numerical results demonstrate the effectiveness of our approach to explain the superiority of QCs over multilayer Perceptron on parity datasets and their limitations over convolutional neural networks on image datasets. Our work sheds light on the problem-dependent power of QNNs and offers a practical tool for evaluating their potential merit.

Association of soluble suppression of tumorigenicity 2 protein with new-onset atrial fibrillation in patients with acute ST-segment elevation myocardial infarction undergoing primary PCI.

Background: Previous studies have indicated that the soluble suppression of tumorigenicity 2 protein (sST2) is associated with new-onset atrial fibrillation (NOAF) in patients diagnosed with coronary artery disease (CAD). However, the predictive value of sST2 in patients with acute ST-segment elevation myocardial infarction (STEMI) undergoing primary percutaneous coronary intervention (PCI) has not been well studied. Methods: A total of 580 patients with STEMI undergoing primary PCI were consecutively recruited between January 2021 and January 2023. These patients were then categorized into two groups: the NOAF group and the no NOAF groups based on the presence of NOAF during admission. The concentration of sST2 in blood samples was measured in all patients. The clinical data from the two groups were prospectively analyzed to investigate the predictive factors of NOAF in patients with STEMI undergoing primary PCI. Results: A total of 41 (7.1%) patients developed NOAF. The presence of NOAF has been found to be associated with various factors, including age, diabetes mellitus, hypertension, the left atrial (LA) diameter, N-terminal pro-brain natriuretic peptide, C-reactive protein (CRP), sST2, a Killip class of ≥2, and a final TIMI flow grade of <3. After including multiple factors, it was observed that LA diameter, CRP, sST2, a Killip class of ≥2, and a final TIMI flow grade of <3 remained significant risk factors for developing NOAF. The receiver operating characteristic (ROC) curve showed the following findings: (1) when the LA diameter exceeded 38.5 mm, the sensitivity and specificity values were observed to be 67.2% and 68.2%, respectively, and the area under the ROC curve (AUC) was 0.683 [95% confidence interval (CI): 0.545-0.732; p = 0.003]; (2) when the CRP level exceeded 8.59, the sensitivity and specificity values were observed to be 68.6% and 69.2%, respectively, and the AUC was 0.713 (95% CI: 0.621-0.778; p < 0.001); and (3) when the sST2 value exceeded 53.3, the sensitivity and specificity values were 79.2% and 68.7%, respectively, and the AUC was 0.799 (95% CI: 0.675-0.865; p < 0.001). Conclusion: sST2 has been identified as an independent predictor of NOAF in patients with STEMI undergoing PCI.

Application of graphene oxide in tumor targeting and tumor therapy.

Graphene oxide (GO), as a kind of two-dimensional sp2 carbon nanomaterials, has attracted great attention in many fields in the past decade. Due to its unique physical and chemical properties, GO is showing great promise in the field of biomedicine. For GO, all the atoms on its surface are exposed to the surface with ultra-high specific surface area, and a variety of groups on the surface, such as carboxyl, hydroxyl and epoxy groups, can effectively bind/load various biomolecules. Due to the availability of these groups, GO also possesses excellent hydrophilicity and biocompatibility for the modification of the desired biocompatible molecules or polymers on the surface of GO. The nano-network structure and hydrophobicity of GO enable it to load a large number of hydrophobic drugs containing benzene rings and it has been widely used as a multi-functional nano-carrier for chemotherapeutic drug or gene delivery. This review article will give an in-depth overview of the synthesis methods of GO, the advantages and disadvantages of GO used in nano-drug delivery system, the research progress of GO as a stimulus-responsive nano-drug carrier, and the application of these intelligent systems in cancer treatment.

Deformable medical image registration with global-local transformation network and region similarity constraint.

Deformable medical image registration can achieve fast and accurate alignment between two images, enabling medical professionals to analyze images of different subjects in a unified anatomical space. As such, it plays an important role in many medical image studies. Current deep learning (DL)-based approaches for image registration directly learn spatial transformation from one image to another, relying on a convolutional neural network and ground truth or similarity metrics. However, these methods only use a global similarity energy function to evaluate the similarity of a pair of images, which ignores the similarity of regions of interest (ROIs) within the images. This can limit the accuracy of the image registration and affect the analysis of specific ROIs. Additionally, DL-based methods often estimate global spatial transformations of images directly, without considering local spatial transformations of ROIs within the images. To address this issue, we propose a novel global-local transformation network with a region similarity constraint that maximizes the similarity of ROIs within the images and estimates both global and local spatial transformations simultaneously. Experiments conducted on four public 3D MRI datasets demonstrate that the proposed method achieves the highest registration performance in terms of accuracy and generalization compared to other state-of-the-art methods.

Femtosecond Laser Preparation of Terracotta Warrior-like Pit Superhydrophobic Structure on Magnesium Alloy with Mechanical Durability and Corrosion Resistance.

Magnesium alloys are among the most promising materials for medical implants, and by preparing a superhydrophobic surface, the rate of corrosion can be effectively slowed down and durability be improved. However, the anticorrosion surfaces are inevitable to be damaged for the conventional micro-nanostructured superhydrophobic magnesium alloys, which highly limits their application prospects. This work proposes the development of a Terracotta Warrior pit superhydrophobic structure (TWPSS), consisting of a wall structure with a Terracotta Warrior-like pit and a lotus-like surface papillae structure within the wall. For the first time, top-down laser ablation of the pits to prepare the lotus-like surface papilla structure is used in conjunction with a bottom-up laser-guided melt stacking of the raised wall structure to achieve rapid fabrication of a TWPSS on a magnesium alloy surface. The Cassie-Baxter-based design of the wall structure spacing effectively protects the internal lotus-like surface papillae from damage and the disappearance of low surface energy material, and the results show that the superhydrophobic surfaces of magnesium alloys have excellent mechanical durability and repairability. In addition, it was found that the recast layer and laser melting stacked layers produced on the surface of the alloy during femtosecond laser processing refined the grain size of the magnesium alloy and effectively suppressed the corrosion rate. The combination of the superhydrophobic gas layer and the resulting grain refinement slowed down the corrosion of the magnesium alloy. Thus, the rapid preparation of TWPSS structures with mechanical durability and corrosion resistance by femtosecond lasers expands the clinical applications of superhydrophobic surface magnesium alloys in medical devices.

Networks of mineral-associated organic matter fractions in forest ecosystems.

Mineral-associated organic matter (MAOM), the largest soil carbon pool, is formed through a series of organo-mineral interaction mechanisms. However, different organo-mineral fractions relevant to specific stabilization mechanisms and their response to environmental variables are poorly understood, which hinders accurate prediction of MAOM preservation under climate change. We applied sequential chemical extraction to separate MAOM into different organo-mineral fractions. To assess of response of different organo-mineral fractions to climate change, alpine forest soils with high environmental sensitivity along a controlled environmental gradient were selected. Residual OM and weakly adsorbed OM were the primary organo-mineral fractions, accounting for approximately 45.1-67.7 % and 16.4-30.6 %, respectively, of the total organic carbon (TOC). Climate exerted considerable indirect effects on the preservation of organo-mineral fractions through weathering and edaphic and biotic variables. Moreover, organo-mineral fractions were closely associated with metal cations (mainly Fe3+/Al3+) and secondary minerals, forming complex networks. Water-soluble OM (WSOM), weakly adsorbed OM and Fe/Al oxyhydroxides-stabilized OM were tightly linked, occupying the central position of the networks, and were closely related to soil pH, moisture and prokaryotic composition, indicating that edaphic and biotic factors might play important roles in maintaining the network structure and topology. In addition, Fe/Al-OM complexes, oxyhydroxides-stabilized OM and residual OM in the network were greatly impacted by climate and weathering factors, including precipitation, temperature and the plagioclase index of alteration (PIA). The complex network among organo-mineral fractions sheds light on MAOM dynamic stabilization for better predicting MAOM preservation under climate change.

pH: A core node of interaction networks among soil organo-mineral fractions.

Mineral-associated organic matter (MAOM) is the largest soil organic carbon (OC) pool with the longest turnover. MAOM is expected to have relatively little sensitivity to climate change due to mineral protection, but its persistence involves several organo-mineral fractions. The uncertainty in the response of specific organo-mineral fractions to climate change hampers the reliability of predictions of MAOM preservation in the future. Here, we applied a sequential chemical fractionation method integrated with network analysis to investigate MAOM stabilization mechanisms across five alpine ecosystems: alpine desert, alpine steppe, alpine meadow, alpine wetland, and alpine forest. Hierarchical cluster analysis revealed grouping of seven extractable OM fractions in MAOM into three OM clusters: a cluster with weak bondings consisting of water-soluble OM (WSOM) and weakly adsorbed fractions (2.1-21.3% of total OC); a cluster with metal-bound complexes comprising Ca-OM complexes and Fe/Al-OM complexes (3.8-12.2% of total OC); and a cluster with strong bonding composed of Al oxyhydroxides, carbonates and Fe oxyhydroxides (12.2-33.5% of total OC). The relative percentages of OM from soils of the five ecosystems in the three clusters exhibited distinct pH dependence patterns. With the increase in pH, the cluster with weak bondings decreased, and that with strong bondings increased, while the one with metal-bound complexes showed a maximum at weakly acidic pH. Organo-mineral fractions and metal cations in MAOM constructed a complex network with pH as the central node. Results suggest that precipitation does not only alter vegetation type and microbial biomass but also regulate soil pH, which is balanced by specific metal cations, thus resulting in particular pH preference of specific OM clusters. These findings demonstrate that soil pH plays a central role in unveiling MAOM dynamics and can serve as a good predictor of soil organo-mineral fractions across alpine ecosystems.

Non-interference delivery of Ce6 and DOX in NIR light-responsive liposomes for synergetic cervical cancer therapy.

Multi-model combination treatment of malignant tumors can make up for the shortcomings of single treatment through multi-target and multi-path to achieve more ideal tumor treatment effect. However, the mutual interference of different drugs in the delivery processin vivoand the difficulty of effective drug accumulation in tumor cells are the bottlenecks of combined therapy. To this project, light-responsive liposomes loading doxorubicin (DOX) and chlorin e6 (Ce6) (DOX-Ce6-Lip) without mutual interference were engineered by thin film hydration method. This kind of nano-drug delivery system increased the drugs concentration accumulated in tumor sites through enhanced permeability and retention effect, and reduced the toxic and side effects of drugs on normal tissuesin vivo. In addition, after entering the tumor cells, Ce6 produced a large number of reactive oxygen species under 660 nm NIR laser irradiation, which further oxidized the unsaturated fatty acid chain in the liposomes and caused the collapse of the liposomes, thus realizing the stimulus-responsive release of Ce6 and DOX. The concentrations of DOX and Ce6 in the tumor cells rapidly reached the peak and achieved a more effective combination of chemotherapy and photodynamic therapy (PDT). Consequently, DOX-Ce6-Lip followed by 660 nm NIR irradiation achieved an efficient tumor growth inhibition of 71.90 ± 3.14%, indicating the versatile potential of chemotherapy and PDT. In conclusion, this study provides a delivery scheme for drugs with different solubilities and an effectively combined anti-tumor therapy method.

Wide field of view optical phased array with a high-directionality antenna.

Light detection and ranging (LiDAR) is a widely utilized technology for extracting information from the outside world in fields such as automotive, robotics, and aerospace. Optical phased array (OPA) is a promising solution for LiDAR technology, although its application is limited by loss and alias-free steering range. In this paper, we propose a dual-layer antenna that achieves a peak directionality of over 92%, thereby mitigating antenna loss and enhancing power efficiency. Based on this antenna, we design and fabricate a 256-channel non-uniform OPA that achieves 150° alias-free steering.

Research Progress of BAP1 in Structure, Function, and Cancer.

Cancer is an important chronic non-communicable disease that endangers human health and has become the main cause of death of residents around the world in the 21st century. At present, most of the mature treatment methods stay at the level of cell and tissue, which is difficult to fundamentally solve the problem of cancer. Therefore, explaining the pathogenesis of cancer at the molecular level becomes the answer to the key problem of cancer regulation. BRCA-associated protein 1 (brca1- associated protein 1) is a kind of ubiquitination enzyme encoded by the BAP1 gene and composed of 729 amino acids. As a carcinogenic protein, BAP1 can affect the cancer cell cycle and proliferation capacity, mutation, and deletion. For example, depending on catalytic activity, it participates in the regulation of intracellular function through transcription, epigenetic, and DNA damage repair. This article mainly reviews the basic structure and function of BAP1 in cells, its role in cancer development, and cancer-related mutants.

Insights From Generative Modeling for Neural Video Compression.

While recent machine learning research has revealed connections between deep generative models such as VAEs and rate-distortion losses used in learned compression, most of this work has focused on images. In a similar spirit, we view recently proposed neural video coding algorithms through the lens of deep autoregressive and latent variable modeling. We present these codecs as instances of a generalized stochastic temporal autoregressive transform, and propose new avenues for further improvements inspired by normalizing flows and structured priors. We propose several architectures that yield state-of-the-art video compression performance on high-resolution video and discuss their tradeoffs and ablations. In particular, we propose (i) improved temporal autoregressive transforms, (ii) improved entropy models with structured and temporal dependencies, and (iii) variable bitrate versions of our algorithms. Since our improvements are compatible with a large class of existing models, we provide further evidence that the generative modeling viewpoint can advance the neural video coding field.