Giant Terahertz Birefringence in an Ultrathin Anisotropic Semimetal.

Manipulating the polarization of light at the nanoscale is key to the development of next-generation optoelectronic devices. This is typically done via waveplates using optically anisotropic crystals, with thicknesses on the order of the wavelength. Here, using a novel ultrafast electron-beam-based technique sensitive to transient near fields at THz frequencies, we observe a giant anisotropy in the linear optical response in the semimetal WTe2 and demonstrate that one can tune the THz polarization using a 50 nm thick film, acting as a broadband wave plate with thickness 3 orders of magnitude smaller than the wavelength. The observed circular deflections of the electron beam are consistent with simulations tracking the trajectory of the electron beam in the near field of the THz pulse. This finding offers a promising approach to enable atomically thin THz polarization control using anisotropic semimetals and defines new approaches for characterizing THz near-field optical response at far-subwavelength length scales.

Three-stage ultrafast demagnetization dynamics in a monolayer ferromagnet.

Intense laser pulses can be used to demagnetize a magnetic material on an extremely short timescale. While this ultrafast demagnetization offers the potential for new magneto-optical devices, it poses challenges in capturing coupled spin-electron and spin-lattice dynamics. In this article, we study the photoinduced ultrafast demagnetization of a prototype monolayer ferromagnet Fe3GeTe2 and resolve the three-stage demagnetization process characterized by an ultrafast and substantial demagnetization on a timescale of 100 fs, followed by light-induced coherent A1g phonon dynamics which is strongly coupled to the spin dynamics in the next 200-800 fs. In the third stage, chiral lattice vibrations driven by nonlinear phonon couplings, both in-plane and out-of-plane are produced, resulting in significant spin precession. Nonadiabatic effects are found to introduce considerable phonon hardening and suppress the spin-lattice couplings during demagnetization. Our results advance our understanding of dynamic charge-spin-lattice couplings in the ultrafast demagnetization and evidence angular momentum transfer between the phonon and spin degrees of freedom.

Engineering Anomalously Large Electron Transport in Topological Semimetals.

Anomalous transport of topological semimetals has generated significant interest for applications in optoelectronics, nanoscale devices, and interconnects. Understanding the origin of novel transport is crucial to engineering the desired material properties, yet their orders of magnitude higher transport than single-particle mobilities remain unexplained. This work demonstrates the dramatic mobility enhancements result from phonons primarily returning momentum to electrons due to phonon-electron dominating over phonon-phonon scattering. Proving this idea, proposed by Peierls in 1932, requires tuning electron and phonon dispersions without changing symmetry, topology, or disorder. This is achieved by combining de Haas - van Alphen (dHvA), electron transport, Raman scattering, and first-principles calculations in the topological semimetals MX2 (M = Nb, Ta and X = Ge, Si). Replacing Ge with Si brings the transport mobilities from an order magnitude larger than single particle ones to nearly balanced. This occurs without changing the crystal structure or topology and with small differences in disorder or Fermi surface. Simultaneously, Raman scattering and first-principles calculations establish phonon-electron dominated scattering only in the MGe2 compounds. Thus, this study proves that phonon-drag is crucial to the transport properties of topological semimetals and provides insight to engineer these materials further.

Enhancing regeneration and repair of long-distance peripheral nerve defect injuries with continuous microcurrent electrical nerve stimulation.

Introduction: Peripheral nerve injuries, especially those involving long-distance deficits, pose significant challenges in clinical repair. This study explores the potential of continuous microcurrent electrical nerve stimulation (cMENS) as an adjunctive strategy to promote regeneration and repair in such cases. Methods: The study initially optimized cMENS parameters and assessed its impact on Schwann cell activity, neurotrophic factor secretion, and the nerve regeneration microenvironment. Subsequently, a rat sciatic nerve defect-bridge repair model was employed to evaluate the reparative effects of cMENS as an adjuvant treatment. Functional recovery was assessed through gait analysis, motor function tests, and nerve conduction assessments. Additionally, nerve regeneration and denervated muscle atrophy were observed through histological examination. Results: The study identified a 10-day regimen of 100uA microcurrent stimulation as optimal. Evaluation focused on Schwann cell activity and the microenvironment, revealing the positive impact of cMENS on maintaining denervated Schwann cell proliferation and enhancing neurotrophic factor secretion. In the rat model of sciatic nerve defect-bridge repair, cMENS demonstrated superior effects compared to control groups, promoting motor function recovery, nerve conduction, and sensory and motor neuron regeneration. Histological examinations revealed enhanced maturation of regenerated nerve fibers and reduced denervated muscle atrophy. Discussion: While cMENS shows promise as an adjuvant treatment for long-distance nerve defects, future research should explore extended stimulation durations and potential synergies with tissue engineering grafts to improve outcomes. This study contributes comprehensive evidence supporting the efficacy of cMENS in enhancing peripheral nerve regeneration.

A multi-dimensional approach to unravel the intricacies of lactylation related signature for prognostic and therapeutic insight in colorectal cancer.

BACKGROUND: Lactylation, a novel contributor to post-translational protein modifications, exhibits dysregulation across various tumors. Nevertheless, its intricate involvement in colorectal carcinoma, particularly for non-histone lactylation and its intersection with metabolism and immune evasion, remains enigmatic. METHODS: Employing immunohistochemistry on tissue microarray with clinical information and immunofluorescence on colorectal cell lines, we investigated the presence of global lactylation and its association with development and progression in colorectal cancer as well as its functional location. Leveraging the AUCell algorithm alongside correlation analysis in single-cell RNA sequencing data, as well as cox-regression and lasso-regression analysis in TCGA dataset and confirmed in GEO dataset, we identified a 23-gene signature predicting colorectal cancer prognosis. Subsequently, we analyzed the associations between the lactylation related gene risk and clinical characteristics, mutation landscapes, biological functions, immune cell infiltration, immunotherapy responses, and drug sensitivity. Core genes were further explored for deep biological insights through bioinformatics and in vitro experiments. RESULTS: Our study innovatively reveals a significant elevation of global lactylation in colorectal cancer, particularly in malignant tumors, confirming it as an independent prognostic factor for CRC. Through a comprehensive analysis integrating tumor tissue arrays, TCGA dataset, GEO dataset, combining in silico investigations and in vitro experiments, we identified a 23-gene Lactylation-Related Gene risk model capable of predicting the prognosis of colorectal cancer patients. Noteworthy variations were observed in clinical characteristics, biological functions, immune cell infiltration, immune checkpoint expression, immunotherapy responses and drug sensitivity among distinct risk groups. CONCLUSIONS: The Lactylation-Related Gene risk model exhibits significant potential for improving the management of colorectal cancer patients and enhancing therapeutic outcomes, particularly at the intersection of metabolism and immune evasion. This finding underscores the clinical relevance of global lactylation in CRC and lays the groundwork for mechanism investigation and targeted therapeutic strategies given the high lactate concentration in CRC.

Skin derived precursors induced Schwann cells mediated tissue engineering-aided neuroregeneration across sciatic nerve defect.

A central question in neural tissue engineering is how the tissue-engineered nerve (TEN) translates detailed transcriptional signals associated with peripheral nerve regeneration into meaningful biological processes. Here, we report a skin-derived precursor-induced Schwann cell (SKP-SC)-mediated chitosan/silk fibroin-fabricated tissue-engineered nerve graft (SKP-SCs-TEN) that can promote sciatic nerve regeneration and functional restoration nearly to the levels achieved by autologous nerve grafts according to behavioral, histological, and electrophysiological evidence. For achieving better effect of neuroregeneration, this is the first time to jointly apply a dynamic perfusion bioreactor and the ascorbic acid to stimulate the SKP-SCs secretion of extracellular matrix (ECM). To overcome the limitation of traditional tissue-engineered nerve grafts, jointly utilizing SKP-SCs and their ECM components were motivated by the thought of prolongating the effect of support cells and their bioactive cues that promote peripheral nerve regeneration. To further explore the regulatory model of gene expression and the related molecular mechanisms involved in tissue engineering-aided peripheral nerve regeneration, we performed a cDNA microarray analysis of gene expression profiling, a comprehensive bioinformatics analysis and a validation study on the grafted segments and dorsal root ganglia tissues. A wealth of transcriptomic and bioinformatics data has revealed complex molecular networks and orchestrated functional regulation that may be responsible for the effects of SKP-SCs-TEN on promoting peripheral nerve regeneration. Our work provides new insights into transcriptomic features and patterns of molecular regulation in nerve functional recovery aided by SKP-SCs-TEN that sheds light on the broader possibilities for novel repair strategies of peripheral nerve injury.

Bckdk-Mediated Branch Chain Amino Acid Metabolism Reprogramming Contributes to Muscle Atrophy during Cancer Cachexia.

SCOPE: Branched chain amino acids (BCAAs) are essential amino acids and important nutrient signals for energy and protein supplementation. The study uses muscle-specific branched-chain α-keto acid dehydrogenase kinase (Bckdk) conditional knockout (cKO) mice to reveal the contribution of BCAA metabolic dysfunction to muscle wasting. METHOD AND RESULTS: Muscle-specific Bckdk-cKO mice are generated through crossbreeding of Bckdkf/f mice with Myf5Cre mice. Lewis lung cancer (LLC) tumor transplantation is used to establish the cancer cachexia model. The occurrence of cancer cachexia is accelerated in the muscle-specific Bckdk-cKO mice after bearing LLC tumor. Wasting skeletal muscle is characterized by increased protein ubiquitination degradation and impaired protein synthesis. The wasting muscle gastrocnemius is mechanized as a distinct BCAA metabolic dysfunction. Based on the atrophy phenotype resulting from BCAA metabolism dysfunction, the optimized BCAA supplementation improves the survival of cancer cachexia in muscle-specific Bckdk-cKO mice bearing LLC tumors, and improves the occurrence of cancer cachexia. The mechanism of BCAA supplementation on muscle mass preservation is based on the promotion of protein synthesis and the inhibition of protein ubiquitination degradation. CONCLUSIONS: Dysfunctional BCAA metabolism contributes to the inhibition of protein synthesis and increases protein degradation in the cancer cachexia model of muscle-specific Bckdk-cKO mice bearing LLC tumors. The reprogramming of BCAA catabolism exerts therapeutic effects by stimulating protein synthesis and inhibiting protein degradation in skeletal muscle.

Unraveling Hidden Charge Density Wave Phases in 1T-TiSe_{2}.

The unexpected chiral order observed in 1T-TiSe_{2} represents an exciting area to explore chirality in condensed matter, while its microscopic mechanism remains elusive. Here, we have identified three metastable collective modes-the so-called single-q modes-in single layer TiSe_{2}, which originate from the unstable phonon eigenvectors at the zone boundary and break the threefold rotational symmetry. We show that polarized laser pulse is a unique and efficient tool to reconstruct the transient potential energy surface, so as to drive phase transitions between these states. By designing sequent layers with chiral stacking order, we propose a practical means to realize chiral charge density waves in 1T-TiSe_{2}. Further, the constructed chiral structure is predicted to exhibit circular dichroism as observed in recent experiments. These facts strongly indicate the chirality transfer from photons to the electron subsystem, meanwhile being strongly coupled to the lattice degree of freedom. Our work provides new insights into understanding and modulating chirality in quantum materials that we hope will spark further experimental investigation.

Oncometabolite D-2-hydroxyglutarate-dependent metabolic reprogramming induces skeletal muscle atrophy during cancer cachexia.

Cancer cachexia is characterized by weight loss and skeletal muscle wasting. Based on the up-regulation of catabolism and down-regulation of anabolism, here we showed genetic mutation-mediated metabolic reprogramming in the progression of cancer cachexia by screening for metabolites and investigating their direct effect on muscle atrophy. Treatment with 93 µM D-2-hydroxyglutarate (D2HG) resulted in reduced myotube width and increased expression of E3 ubiquitin ligases. Isocitrate Dehydrogenase 1 (IDH1) mutant patients had higher D2HG than non-mutant patients. In the in vivo murine cancer cachexia model, mutant IDH1 in CT26 cancer cells accelerated cachexia progression and worsened overall survival. Transcriptomics and metabolomics revealed a distinct D2HG-induced metabolic imbalance. Treatment with the IDH1 inhibitor ivosidenib delayed the progression of cancer cachexia in murine GL261 glioma model and CT26 colorectal carcinoma models. These data demonstrate the contribution of IDH1 mutation mediated D2HG accumulation to the progression of cancer cachexia and highlight the individualized treatment of IDH1 mutation associated cancer cachexia.

Hprt Serves as an Ideal Reference Gene for qRT-PCR Normalization in Rat DRG Neurons.

OBJECTIVE: To identify suitable reference genes for gene expression studies in rat dorsal root ganglia (DRG) neurons. METHODS: The raw cycle threshold (Ct) values of 12 selected reference genes were obtained via quantitative real-time reverse transcription polymerase chain reaction (qRT-PCR) in neurons at different developmental stages or under different treatments. Two strategies were employed to screen the most stable reference genes: the genes were ranked according to the coefficient of biological variation and further validated using geNorm and NormFinder programs. The stable and unstable reference genes were subsequently used as internal controls to assess their effects on target gene expression. RESULTS: All reference genes showed varying degrees of fluctuation in Ct values during the growth process of neurons or after different treatments. 18S ribosomal RNA (Rn18s) and ß-actin (Actb) exhibited the most significant changes, while ubiquitin C (Ubc), hypoxanthine phosphoribosyl transferase (Hprt), and mitochondrial ribosomal protein L10 (Mrpl10) showed relatively minor changes. The most stable and unstable genes obtained by different evaluation methods varied slightly. Overall, Actb was found to be the most unstable reference gene, while Hprt was the relatively most stable reference gene. The use of unstable reference genes Actb and ankyrin repeat domain 27 (Ankrd27) as internal controls led to high variability within the control group, ultimately affecting the determination of target gene expression. In contrast, the stable reference gene Hprt had small inter-assay variation and high stability. CONCLUSIONS: Our observations indicate that Hprt is a proper endogenous reference gene for qRT-PCR analysis in rat DRG neurons and thus provides a critical molecular basis for the genetic characterization in neurological disorders.

AF9 sustains glycolysis in colorectal cancer via H3K9ac-mediated PCK2 and FBP1 transcription.

BACKGROUND: The tumourigenesis of various cancers is influenced by epigenetic deregulation. Among 591 epigenetic regulator factors (ERFs) examined, AF9 showed significant inhibition of malignancy in colorectal cancer (CRC) based on our wound healing assays. However, the precise role of AF9 in CRC remains to be explored. METHODS: To investigate the function of AF9 in CRC, we utilised small interfering RNAs (siRNAs) to knock down the expression of 591 ERFs. Subsequently, we performed wound healing assays to evaluate cell proliferation and migration. In vitro and in vivo assays were conducted to elucidate the potential impact of AF9 in CRC. Clinical samples were analysed to assess the association between AF9 expression and CRC prognosis. Additionally, an Azoxymethane-Dextran Sodium Sulfate (AOM/DSS) induced CRC AF9IEC-/- mouse model was employed to confirm the role of AF9 in CRC. To identify the target gene of AF9, RNA-seq and coimmunoprecipitation analyses were performed. Furthermore, bioinformatics prediction was applied to identify potential miRNAs that target AF9. RESULTS: Among the 591 ERFs examined, AF9 exhibited downregulation in CRC and showed a positive correlation with prolonged survival in CRC patients. In vitro and in vivo assays proved that depletion of AF9 could promote cell proliferation, migration as well as glycolysis. Specifically, knockout of MLLT3 (AF9) in intestinal epithelial cells significantly increased tumour formation induced by AOM/DSS. We also identified miR-145 could target 3'untranslated region of AF9 to suppress AF9 expression. Loss of AF9 led to decreased expression of gluconeogenic genes, including phosphoenolpyruvate carboxykinase 2 (PCK2) and fructose 1,6-bisphosphatase 1 (FBP1), subsequently promoting glucose consumption and tumourigenesis. CONCLUSIONS: AF9 is essential for the upregulation of PCK2 and FBP1, and the disruption of the miR-145/AF9 axis may serve as a potential target for the development of CRC therapeutics.

Coherent-Phonon-Driven Intervalley Scattering and Rabi Oscillation in Multivalley 2D Materials.

Resolving the complete electron scattering dynamics mediated by coherent phonons is crucial for understanding electron-phonon couplings beyond equilibrium. Here we present a time-resolved theoretical investigation on strongly coupled ultrafast electron and phonon dynamics in monolayer WSe_{2}, with a focus on the intervalley scattering from the optically "bright" K state to "dark" Q state. We find that the strong coherent lattice vibration along the longitudinal acoustic phonon mode [LA(M)] can drastically promote K-to-Q transition on a timescale of ∼400 fs, comparable with previous experimental observation on thermal-phonon-mediated electron dynamics. Further, this coherent-phonon-driven intervalley scattering occurs in an unconventional steplike manner and further induces an electronic Rabi oscillation. By constructing a two-level model and quantitatively comparing with ab initio dynamic simulations, we uncover the critical role of nonadiabatic coupling effects. Finally, a new strategy is proposed to effectively tune the intervalley scattering rates by varying the coherent phonon amplitude, which could be realized via light-induced nonlinear phononics that we hope will spark experimental investigation.

DisAVR: Disentangled Adaptive Visual Reasoning Network for Diagram Question Answering.

Diagram Question Answering (DQA) aims to correctly answer questions about given diagrams, which demands an interplay of good diagram understanding and effective reasoning. However, the same appearance of objects in diagrams can express different semantics. This kind of visual semantic ambiguity problem makes it challenging to represent diagrams sufficiently for better understanding. Moreover, since there are questions about diagrams from different perspectives, it is also crucial to perform flexible and adaptive reasoning on content-rich diagrams. In this paper, we propose a Disentangled Adaptive Visual Reasoning Network for DQA, named DisAVR, to jointly optimize the dual-process of representation and reasoning. DisAVR mainly comprises three modules: improved region feature learning, question parsing, and disentangled adaptive reasoning. Specifically, the improved region feature learning module is designed to first learn robust diagram representation by integrating detail-aware patch features and semantically-explicit text features with region features. Subsequently, the question parsing module decomposes the question into three types of question guidance including region, spatial relation and semantic relation guidance to dynamically guide subsequent reasoning. Next, the disentangled adaptive reasoning module decomposes the whole reasoning process by employing three visual reasoning cells to construct a soft fully-connected multi-layer stacked routing space. These three cells in each layer reason over object regions, semantic and spatial relations in the diagram under the corresponding question guidance. Moreover, an adaptive routing mechanism is designed to flexibly explore more optimal reasoning paths for specific diagram-question pairs. Extensive experiments on three DQA datasets demonstrate the superiority of our DisAVR.

Liver metastasis from colorectal cancer: pathogenetic development, immune landscape of the tumour microenvironment and therapeutic approaches.

Colorectal cancer liver metastasis (CRLM) is one of the leading causes of death among patients with colorectal cancer (CRC). Although immunotherapy has demonstrated encouraging outcomes in CRC, its benefits are minimal in CRLM. The complex immune landscape of the hepatic tumour microenvironment is essential for the development of a premetastatic niche and for the colonisation and metastasis of CRC cells; thus, an in-depth understanding of these mechanisms can provide effective immunotherapeutic targets for CRLM. This review summarises recent studies on the immune landscape of the tumour microenvironment of CRLM and highlights therapeutic prospects for targeting the suppressive immune microenvironment of CRLM.

Cucurbitacin IIb attenuates cancer cachexia induced skeletal muscle atrophy by regulating the IL-6/STAT3/FoxO signaling pathway.

The main features of cancer cachexia include skeletal muscle atrophy, which can significantly reduce the quality of life of patients. Clinical treatment of cancer cachexia is mainly based on nutritional therapy and physical exercise; medication only improves appetite but does not reverse the symptoms of skeletal muscle wasting. In this work, we systematically studied the underlying molecular mechanisms by which cucurbitacin IIb (CuIIb) ameliorates muscle wasting in cancer cachexia both in vitro and in vivo. CuIIb significantly ameliorated the chief features of cancer cachexia in vivo, alleviating weight loss, food intake, muscle wasting, adipose tissue depletion, and organ weight reductions. In vitro, CuIIb (10 and 20 µM) dose-dependently attenuated conditioned medium (CM)-induced C2C12 myotube atrophy. Collectively, our findings demonstrated that CuIIb prevented the upregulation of the E3 ubiquitin ligase muscle atrophy Fbox protein (MAFbx), myosin heavy chain (MyHC), and myogenin (MyoG) and impacted protein synthesis and degradation. In addition, CuIIb decreased the phosphorylation of Tyr705 in STAT3 by regulating the IL-6/STAT3/FoxO pathway to reduce skeletal muscle atrophy in cancer cachexia.

Downregulation of UBE4B promotes CNS axon regrowth and functional recovery after stroke.

The limited intrinsic regrowth capacity of corticospinal axons impedes functional recovery after cortical stroke. Although the mammalian target of rapamycin (mTOR) and p53 pathways have been identified as the key intrinsic pathways regulating CNS axon regrowth, little is known about the key upstream regulatory mechanism by which these two major pathways control CNS axon regrowth. By screening genes that regulate ubiquitin-mediated degradation of the p53 proteins in mice, we found that ubiquitination factor E4B (UBE4B) represses axonal regrowth in retinal ganglion cells and corticospinal neurons. We found that axonal regrowth induced by UBE4B depletion depended on the cooperative activation of p53 and mTOR. Importantly, overexpression of UbV.E4B, a competitive inhibitor of UBE4B, in corticospinal neurons promoted corticospinal axon sprouting and facilitated the recovery of corticospinal axon-dependent function in a cortical stroke model. Thus, our findings provide a translatable strategy for restoring corticospinal tract-dependent functions after cortical stroke.

BST2 induced macrophage M2 polarization to promote the progression of colorectal cancer.

Background: Tumor-associated macrophages (TAMs) are one of the most prominent tumor-infiltrating immune cells in the tumor microenvironment (TME) of CRC and play a vital role in the progression of CRC. BST2 was predicted to be associated with the infiltration of TAMs. However, its potential function by which CRC cells and TAMs interact with each other still needs further investigation. Methods: The target genes in CRC were selected by bioinformatics screening. The level of bone marrow stromal cell antigen 2 (BST2) in CRC cells and tissues was determined by qRT‒PCR, Western blotting, and immunohistochemistry staining. In vitro and in vivo assays were applied to clarify the function of BST2. Results: In this study, according to bioinformatics analysis, a nomogram based on the risk score (constructed by BST2 and CAV1 (caveolin-1)) and clinical features was built and displayed satisfactory prognostic value. Upregulated BST2 was significantly related to Braf mutation, dMMR/MSI-H, CMS1 subtype, and immune response and was a potential biomarker for predicting immune checkpoint blockade therapy. Silencing BST2 in CRC obviously restrained CRC progression and M2 TAM polarization. The infiltration of TAMs was positively correlated with the high expression of BST2, and depletion of TAMs alleviated the protumoural effect of BST2 in CRC in vivo. In vitro experiments revealed that a reduction in BST2 in CRC inhibited CRC proliferation and migration and also M2 polarization. Conclusion: These findings indicated that BST2 played a vital role in CRC progression and might be a predictable marker for immunotherapy.

Scutellarin Attenuates Doxorubicin-Induced Cardiotoxicity by Inhibiting Myocardial Fibrosis, Apoptosis and Autophagy in Rats.

The anthracycline antibiotic doxorubicin (DOX) is an effective anticancer agent, but its clinical use is limited by dose-dependent cardiotoxicity. Scutellarin (SCU), a natural polyphenolic flavonoid, is used as a cardioprotective agent for infarction and ischemia-reperfusion injury. This study investigated the beneficial effect of SCU on DOX-induced chronic cardiotoxicity. Rats were injected intraperitoneally (i. p.) with DOX (2.5 mg/kg) twice a week for four weeks and then allowed to rest for two weeks to establish the chronic cardiotoxicity animal model. A dose of 10 mg/kg/day SCU was injected i. p. daily for six weeks to attenuate cardiotoxicity. SCU attenuated DOX-induced elevated oxidative stress levels and cardiac troponin T (cTnT), decreased left ventricular ejection fraction (LVEF) and fractional shortening (LVFS), elevated isovolumic relaxation time (IVRT), electrophysiology and histopathological alterations. In addition, SCU significantly attenuated DOX-induced cardiac fibrosis and reduced extracellular matrix (ECM) accumulation by inhibiting the TGF-ß1/Smad2 signaling pathway. Furthermore, SCU also prevented against DOX-induced apoptosis and autophagy as evidenced by upregulation of Bcl-2, downregulation of Bax and cleaved caspase-3, inhibited the AMPK/mTOR pathway. These results revealed that the cardioprotective effect of SCU on DOX-induced chronic cardiotoxicity may be attributed to reducing oxidative stress, myocardial fibrosis, apoptosis and autophagy.

Multitask Learning for Visual Question Answering.

Visual question answering (VQA) is a task that machines should provide an accurate natural language answer given an image and a question about the image. Many studies have found that the current VQA methods are heavily driven by the surface correlation or statistical bias in the training data, and lack sufficient image grounding. To address this issue, we devise a novel end-to-end architecture that uses multitask learning to promote more sufficient image grounding and learn effective multimodality representations. The tasks consist of VQA and our proposed image cloze (IC) task requires machines to fill in the blanks accurately given an image and a textual description of the image. To ensure our model performs sufficient image grounding as much as possible, we propose a novel word-masking algorithm to develop the multimodal IC task based on the part-of-speech of words. Our model predicts the VQA answer and fills in the blanks after the multimodality representation learning that is shared by the two tasks. Experimental results show that our model achieves almost the equivalent, state-of-the-art, second-best performance on the VQA v2.0, VQA-changing priors (CP) v2, and grounded question answering (GQA) datasets, respectively, with fewer parameters and without additional data compared with baselines.

Relation-Aware Fine-Grained Reasoning Network for Textbook Question Answering.

Textbook question answering (TQA) is a task that one should answer non-diagram and diagram questions accurately, given a large context which consists of abundant diagrams and essays. Although lots of studies have made significant progress in the natural image question answering (QA), they are not applicable to comprehending diagrams and reasoning over the long multimodal context. To address the above issues, we propose a relation-aware fine-grained reasoning (RAFR) network that performs fine-grained reasoning over the nodes of relation-based diagram graphs. Our method uses semantic dependencies and relative positions between nodes in the diagram to construct relation graphs and applies graph attention networks to learn diagram representations. To extract and reason over the multimodal knowledge, we first extract the text that is the most relevant to questions, options, and the instructional diagram which is the most relevant to question diagrams at the word-sentence level and the node-diagram level, respectively. Then, we apply instructional-diagram-guided attention and question-guided attention to reason over the node of question diagrams, respectively. The experimental results show that our proposed method achieves the best performance on the TQA dataset compared with baselines. We also conduct extensive ablation studies to comprehensively analyze the proposed method.