Recent advances in discrete diffusion models have demonstrated strong performance in image generation and masked language modeling, yet they remain limited in their capacity for controlled content editing. We propose DICE (Discrete Inversion for Controllable Editing), a novel framework that pioneers precise inversion capabilities for discrete diffusion models, including both masked generative and multinomial diffusion variants. Our key innovation lies in capturing noise sequences and masking patterns during reverse diffusion process, enabling both accurate reconstruction and flexible editing without relying on predefined masks or attention-based manipulations. Through comprehensive experiments across image and text modalities using models such as Paella, VQ-Diffusion, RoBERTa and LLaDA, we demonstrate that DICE successfully maintains high fidelity to the original data while significantly expanding editing capabilities. These results establish new possibilities for fine-grained content manipulation in discrete spaces.

Vision Transformers (ViTs) have achieved state-of-the-art performance in offline video action detection, but their reliance on processing fixed-size clips with full spatiotemporal attention makes them computationally expensive and ill-suited for real-time streaming applications due to massive computational redundancy. This paper introduces a novel framework to adapt these powerful offline models into efficient, online student models through knowledge distillation. We propose two causal, streaming-friendly attention architectures that replace the full self-attention mechanism: (1) a lightweight Temporal Shift Attention that integrates past context with minimal overhead, and (2) a Decomposed Spatial-Temporal Attention that combines intraframe spatial attention with an LSTM for temporal modeling. Both architectures utilize caching to eliminate redundant operations on a frame-by-frame basis. To maximize knowledge transfer, we introduce an uncertainty-guided distillation process, which formulates the training as a multi-task learning problem. Our resulting models demonstrate significant efficiency gains, achieving up to a 4x improvement in latency and throughput compared to the original offline teacher while ensuring state-of-the-art performance for online methods. Our work provides a practical and effective methodology for deploying high-accuracy transformer models in latency-sensitive, real-world video analysis systems.

We consider the conditional generation of 3D drug-like molecules with explicit control over molecular properties such as drug-like properties (e.g., Quantitative Estimate of Druglikeness or Synthetic Accessibility score) and effectively binding to specific protein sites. To tackle this problem, we propose an E(3)-equivariant Wasserstein autoencoder and factorize the latent space of our generative model into two disentangled aspects: molecular properties and the remaining structural context of 3D molecules. Our model ensures explicit control over these molecular attributes while maintaining equivariance of coordinate representation and invariance of data likelihood. Furthermore, we introduce a novel alignment-based coordinate loss to adapt equivariant networks for auto-regressive de-novo 3D molecule generation from scratch. Extensive experiments validate our model's effectiveness on property-guided and context-guided molecule generation, both for de-novo 3D molecule design and structure-based drug discovery against protein targets.

Designing protein-binding proteins with high affinity is critical in biomedical research and biotechnology. Despite recent advancements targeting specific proteins, the ability to create high-affinity binders for arbitrary protein targets on demand, without extensive rounds of wet-lab testing, remains a significant challenge. Here, we introduce PPDiff, a diffusion model to jointly design the sequence and structure of binders for arbitrary protein targets in a non-autoregressive manner. PPDiff builds upon our developed Sequence Structure Interleaving Network with Causal attention layers (SSINC), which integrates interleaved self-attention layers to capture global amino acid correlations, $k$-nearest neighbor ($k$NN) equivariant graph convolutional layers to model local interactions in three-dimensional (3D) space, and causal attention layers to simplify the intricate interdependencies within the protein sequence. To assess PPDiff, we curate PPBench, a general protein-protein complex dataset comprising 706,360 complexes from the Protein Data Bank (PDB). The model is pretrained on PPBench and finetuned on two real-world applications: target-protein mini-binder complex design and antigen-antibody complex design. PPDiff consistently surpasses baseline methods, achieving success rates of 50.00\%, 23.16\%, and 16.89\% for the pretraining task and the two downstream applications, respectively.

Cell fate decisions are highly coordinated processes governed by complex interactions among numerous regulatory genes, while disruptions in these mechanisms can lead to developmental abnormalities and disease. Traditional methods often fail to capture such combinatorial interactions, limiting their ability to fully model cell fate dynamics. Here, we introduce MetaVelo, a global feature explanation framework for identifying key regulatory gene sets influencing cell fate transitions. MetaVelo models these transitions as a black-box function and employs a differentiable neural ordinary differential equation (ODE) surrogate to enable efficient optimization. By reparameterizing the problem as a controllable data generation process, MetaVelo overcomes the challenges posed by the non-differentiable nature of cell fate dynamics. Benchmarking across diverse stand-alone and longitudinal single-cell RNA-seq datasets and three black-box cell fate models demonstrates its superiority over 12 baseline methods in predicting developmental trajectories and identifying combinatorial regulatory gene sets. MetaVelo further distinguishes independent from synergistic regulatory genes, offering novel insights into the gene interactions governing cell fate. With the growing availability of high-resolution single-cell data, MetaVelo provides a scalable and effective framework for advancing developmental biology and therapeutic applications.

Action detection aims to detect (recognize and localize) human actions spatially and temporally in videos. Existing approaches focus on the closed-set setting where an action detector is trained and tested on videos from a fixed set of action categories. However, this constrained setting is not viable in an open world where test videos inevitably come beyond the trained action categories. In this paper, we address the practical yet challenging Open-Vocabulary Action Detection (OVAD) problem. It aims to detect any action in test videos while training a model on a fixed set of action categories. To achieve such an open-vocabulary capability, we propose a novel method OpenMixer that exploits the inherent semantics and localizability of large vision-language models (VLM) within the family of query-based detection transformers (DETR). Specifically, the OpenMixer is developed by spatial and temporal OpertMixer blocks (S-OMB and T-OMB), and a dynamically fused alignment (DFA) module. The three components collectively enjoy the merits of strong generalization from pretrained VLMs and end-to-end learning from DETR design. Moreover, we established OVAD benchmarks under various settings, and the experimental results show that the OpenMixer performs the best over baselines for detecting seen and unseen actions. We release the codes, models, and dataset splits at https://github.com/Cogito2012/0penMixer.

Abstract Despite the recent impressive breakthroughs in text-to-image generation, generative models have difficulty in capturing the data distribution of underrepresented attribute compositions while over-memorizing overrepresented attribute compositions, which raises public concerns about their robustness and fairness. To tackle this challenge, we propose ACTIG, an attribute-centric compositional text-to-image generation framework. We present an attribute-centric feature augmentation and a novel image-free training scheme, which greatly improves model’s ability to generate images with underrepresented attributes. We further propose an attribute-centric contrastive loss to avoid overfitting to overrepresented attribute compositions. We validate our framework on the CelebA-HQ and CUB datasets. Extensive experiments show that the compositional generalization of ACTIG is outstanding, and our framework outperforms previous works in terms of image quality and text-image consistency. The source code and trained models are publicly available at https://github.com/yrcong/ACTIG .

In this paper, we explore the capability of an agent to construct a logical sequence of action steps, thereby as-sembling a strategic procedural plan. This plan is cru-cial for navigating from an initial visual observation to a target visual outcome, as depicted in real-life instructional videos. Existing works have attained partial suc-cess by extensively leveraging various sources of information available in the datasets, such as heavy intermediate visual observations, procedural names, or natural language step-by-step instructions, for features or supervision signals. However, the task remains formidable due to the implicit causal constraints in the sequencing of steps and the variability inherent in multiple feasible plans. To tackle these intricacies that previous efforts have over-looked, we propose to enhance the agent's capabilities by infusing it with procedural knowledge. This knowledge, sourced from training procedure plans and structured as a directed weighted graph, equips the agent to better nav-igate the complexities of step sequencing and its poten-tial variations. We coin our approach KEPP, a novel Knowledge-Enhanced Procedure Planning system, which harnesses a probabilistic procedural knowledge graph ex-tracted from training data, effectively acting as a compre-hensive textbook for the training domain. Experimental evaluations across three widely-used datasets under set-tings of varying complexity reveal that KEPP attains su-perior, state-of-the-art results while requiring only mini-mal supervision. Code and trained model are available at https://github.com/Ravindu-Yasas-Nagasinghe/KEPP

When performing complex multi-step reasoning tasks, the ability of Large Language Models (LLMs) to derive structured intermediate proof steps is important for ensuring that the models truly perform the desired reasoning and for improving models' explainability.This paper is centred around a focused study: whether the current state-of-the-art generalist LLMs can leverage the structures in a few examples to better construct the proof structures with incontext learning.Our study specifically focuses on structure-aware demonstration and structureaware pruning.We demonstrate that they both help improve performance.A detailed analysis is provided to help understand the results. 1

MOTIVATION: Recent advances in spatial transcriptomics allow spatially resolved gene expression measurements with cellular or even sub-cellular resolution, directly characterizing the complex spatiotemporal gene expression landscape and cell-to-cell interactions in their native microenvironments. Due to technology limitations, most spatial transcriptomic technologies still yield incomplete expression measurements with excessive missing values. Therefore, gene imputation is critical to filling in missing data, enhancing resolution, and improving overall interpretability. However, existing methods either require additional matched single-cell RNA-seq data, which is rarely available, or ignore spatial proximity or expression similarity information. RESULTS: To address these issues, we introduce Impeller, a path-based heterogeneous graph learning method for spatial transcriptomic data imputation. Impeller has two unique characteristics distinct from existing approaches. First, it builds a heterogeneous graph with two types of edges representing spatial proximity and expression similarity. Therefore, Impeller can simultaneously model smooth gene expression changes across spatial dimensions and capture similar gene expression signatures of faraway cells from the same type. Moreover, Impeller incorporates both short- and long-range cell-to-cell interactions (e.g. via paracrine and endocrine) by stacking multiple GNN layers. We use a learnable path operator in Impeller to avoid the over-smoothing issue of the traditional Laplacian matrices. Extensive experiments on diverse datasets from three popular platforms and two species demonstrate the superiority of Impeller over various state-of-the-art imputation methods. AVAILABILITY AND IMPLEMENTATION: The code and preprocessed data used in this study are available at https://github.com/aicb-ZhangLabs/Impeller and https://zenodo.org/records/11212604.

MOTIVATION: Spatial transcriptomics technologies, which generate a spatial map of gene activity, can deepen the understanding of tissue architecture and its molecular underpinnings in health and disease. However, the high cost makes these technologies difficult to use in practice. Histological images co-registered with targeted tissues are more affordable and routinely generated in many research and clinical studies. Hence, predicting spatial gene expression from the morphological clues embedded in tissue histological images provides a scalable alternative approach to decoding tissue complexity. RESULTS: Here, we present a graph neural network based framework to predict the spatial expression of highly expressed genes from tissue histological images. Extensive experiments on two separate breast cancer data cohorts demonstrate that our method improves the prediction performance compared to the state-of-the-art, and that our model can be used to better delineate spatial domains of biological interest. AVAILABILITY AND IMPLEMENTATION: https://github.com/song0309/asGNN/.

Source-free domain adaptation (SFDA) is an emerging research topic that studies how to adapt a pretrained source model using unlabeled target data. It is derived from unsupervised domain adaptation but has the advantage of not requiring labeled source data to learn adaptive models. This makes it particularly useful in real-world applications where access to source data is restricted. While there has been some SFDA work for images, little attention has been paid to videos. Naively extending image-based methods to videos without considering the unique properties of videos often leads to unsatisfactory results. In this paper, we propose a simple and highly flexible method for Source-Free Video Domain Adaptation (SFVDA), which extensively exploits consistency learning for videos from spatial, temporal, and historical perspectives. Our method is based on the assumption that videos of the same action category are drawn from the same low-dimensional space, regardless of the spatio-temporal variations in the high-dimensional space that cause domain shifts. To overcome domain shifts, we simulate spatio-temporal variations by applying spatial and temporal augmentations on target videos and encourage the model to make consistent predictions from a video and its augmented versions. Due to the simple design, our method can be applied to various SFVDA settings, and experiments show that our method achieves state-of-the-art performance for all the settings.

Conditional image-to-video (cI2V) generation aims to synthesize a new plausible video starting from an image (e.g., a person's face) and a condition (e.g., an action class label like smile). The key challenge of the cI2V task lies in the simultaneous generation of realistic spatial appearance and temporal dynamics corresponding to the given image and condition. In this paper, we propose an approach for cI2V using novel latent flow diffusion models (LFDM) that synthesize an optical flow sequence in the latent space based on the given condition to warp the given image. Compared to previous direct-synthesis-based works, our proposed LFDM can better synthesize spatial details and temporal motion by fully utilizing the spatial content of the given image and warping it in the latent space according to the generated temporally-coherent flow. The training of LFDM consists of two separate stages: (1) an unsupervised learning stage to train a latent flow auto-encoder for spatial content generation, including a flow predictor to estimate latent flow between pairs of video frames, and (2) a conditional learning stage to train a 3D-UNet-based diffusion model (DM) for temporal latent flow generation. Unlike previous DMs operating in pixel space or latent feature space that couples spatial and temporal information, the DM in our LFDM only needs to learn a low-dimensional latent flow space for motion generation, thus being more computationally efficient. We conduct comprehensive experiments on multiple datasets, where LFDM consistently outperforms prior arts. Furthermore, we show that LFDM can be easily adapted to new domains by simply finetuning the image decoder. Our code is available at https://github.com/nihaomiao/CVPR23_LFDM.

Recent few-shot video classification (FSVC) works achieve promising performance by capturing similarity across support and query samples with different temporal alignment strategies or learning discriminative features via Transformer block within each episode. However, they ignore two important issues: a) It is difficult to capture rich intrinsic action semantics from a limited number of support instances within each task. b) Redundant or irrelevant frames in videos easily weaken the positive influence of discriminative frames. To address these two issues, this paper proposes a novel Representation Fusion and Promotion Learning (RFPL) mechanism with two sub-modules: meta-action learning (MAL) and reinforced image representation (RIR). Concretely, during training stage, we perform online learning for seeking a task-shared meta-action bank to enrich task-specific action representation by injecting global knowledge. Besides, we exploit reinforcement learning to obtain the importance of each frame and refine the representation. This operation maximizes the contribution of discriminative frames to further capture the similarity of support and query samples from the same category. Our RFPL framework is highly flexible that it can be integrated with many existing FSVC methods. Extensive experiments show that RFPL significantly enhances the performance of existing FSVC models when integrated with them.

In protein biophysics, the separation between the functionally important residues (forming the active site or binding surface) and those that create the overall structure (the fold) is a well-established and fundamental concept. Identifying and modifying those functional sites is critical for protein engineering but computationally non-trivial, and requires significant domain knowledge. To automate this process from a data-driven perspective, we propose a disentangled Wasserstein autoencoder with an auxiliary classifier, which isolates the function-related patterns from the rest with theoretical guarantees. This enables one-pass protein sequence editing and improves the understanding of the resulting sequences and editing actions involved. To demonstrate its effectiveness, we apply it to T-cell receptors (TCRs), a well-studied structure-function case. We show that our method can be used to alter the function of TCRs without changing the structural backbone, outperforming several competing methods in generation quality and efficiency, and requiring only 10\% of the running time needed by baseline models. To our knowledge, this is the first approach that utilizes disentangled representations for TCR engineering.

MOTIVATION: MHC Class I protein plays an important role in immunotherapy by presenting immunogenic peptides to anti-tumor immune cells. The repertoires of peptides for various MHC Class I proteins are distinct, which can be reflected by their diverse binding motifs. To characterize binding motifs for MHC Class I proteins, in vitro experiments have been conducted to screen peptides with high binding affinities to hundreds of given MHC Class I proteins. However, considering tens of thousands of known MHC Class I proteins, conducting in vitro experiments for extensive MHC proteins is infeasible, and thus a more efficient and scalable way to characterize binding motifs is needed. RESULTS: We presented a de novo generation framework, coined PepPPO, to characterize binding motif for any given MHC Class I proteins via generating repertoires of peptides presented by them. PepPPO leverages a reinforcement learning agent with a mutation policy to mutate random input peptides into positive presented ones. Using PepPPO, we characterized binding motifs for around 10 000 known human MHC Class I proteins with and without experimental data. These computed motifs demonstrated high similarities with those derived from experimental data. In addition, we found that the motifs could be used for the rapid screening of neoantigens at a much lower time cost than previous deep-learning methods. AVAILABILITY AND IMPLEMENTATION: The software can be found in https://github.com/minrq/pMHC. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

MOTIVATION: We present a multi-sequence generalization of Variational Information Bottleneck and call the resulting model Attentive Variational Information Bottleneck (AVIB). Our AVIB model leverages multi-head self-attention to implicitly approximate a posterior distribution over latent encodings conditioned on multiple input sequences. We apply AVIB to a fundamental immuno-oncology problem: predicting the interactions between T-cell receptors (TCRs) and peptides. RESULTS: Experimental results on various datasets show that AVIB significantly outperforms state-of-the-art methods for TCR-peptide interaction prediction. Additionally, we show that the latent posterior distribution learned by AVIB is particularly effective for the unsupervised detection of out-of-distribution amino acid sequences. AVAILABILITY AND IMPLEMENTATION: The code and the data used for this study are publicly available at: https://github.com/nec-research/vibtcr. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Although progress has been made for text-to-image synthesis, previous methods fall short of generalizing to unseen or underrepresented attribute compositions in the input text. Lacking compositionality could have severe implications for robustness and fairness, e.g., inability to synthesize the face images of underrepresented demographic groups. In this paper, we introduce a new framework, StyleT2I, to improve the compositionality of text-to-image synthesis. Specifically, we propose a CLIP-guided Contrastive Loss to better distinguish different compositions among different sentences. To further improve the compositionality, we design a novel Semantic Matching Loss and a Spatial Constraint to identify attributes' latent directions for intended spatial region manipulations, leading to better disentangled latent representations of attributes. Based on the identified latent directions of attributes, we propose Compositional Attribute Adjustment to adjust the latent code, resulting in better compositionality of image synthesis. In addition, we leverage the$l_{2}$-norm regularization of identified latent directions (norm penalty) to strike a nice balance between image-text alignment and image fidelity. In the experiments, we devise a new dataset split and an evaluation metric to evaluate the compositionality of text-to-image synthesis models. The results show that StyleT2I outperforms previous approaches in terms of the consistency between the input text and synthesized images and achieves higher fidelity.

Predicting the interactions between T-cell receptors (TCRs) and peptides is crucial for the development of personalized medicine and targeted vaccine in immunotherapy. Current datasets for training deep learning models of this purpose remain constrained without diverse TCRs and peptides. To combat the data scarcity issue presented in the current datasets, we propose to extend the training dataset by physical modeling of TCR-peptide pairs. Specifically, we compute the docking energies between auxiliary unknown TCR-peptide pairs as surrogate training labels. Then, we use these extended example-label pairs to train our model in a supervised fashion. Finally, we find that the AUC score for the prediction of the model can be further improved by pseudo-labeling of such unknown TCR-peptide pairs (by a trained teacher model), and re-training the model with those pseudo-labeled TCR-peptide pairs. Our proposed method that trains the deep neural network with physical modeling and data-augmented pseudo-labeling improves over baselines in the available two datasets. We also introduce a new dataset that contains over 80,000 unknown TCR-peptide pairs with docking energy scores.

StyleGANs have shown impressive results on data generation and manipulation in recent years, thanks to its disentangled style latent space. A lot of efforts have been made in inverting a pretrained generator, where an encoder is trained ad hoc after the generator is trained in a two-stage fashion. In this paper, we focus on style-based generators asking a scientific question: Does forcing such a generator to reconstruct real data lead to more disentangled latent space and make the inversion process from image to latent space easy? We describe a new methodology to train a style-based autoencoder where the encoder and generator are optimized end-to-end. We show that our proposed model consistently outperforms baselines in terms of image inversion and generation quality. Supplementary, code, and pretrained models are available on the project website1.

Conditional Generative Adversarial Networks (cGANs) extend the standard unconditional GAN framework to learning joint data-label distributions from samples, and have been established as powerful generative models capable of generating high-fidelity imagery. A challenge of training such a model lies in properly infusing class information into its generator and discriminator. For the discriminator, class conditioning can be achieved by either (1) directly incorporating labels as input or (2) involving labels in an auxiliary classification loss. In this paper, we show that the former directly aligns the class-conditioned fake-and-real data distributions P (image|class) (data matching), while the latter aligns data-conditioned class distributions P (class|image) (label matching). Although class separability does not directly translate to sample quality and becomes a burden if classification itself is intrinsically difficult, the discriminator cannot provide useful guidance for the generator if features of distinct classes are mapped to the same point and thus become inseparable. Motivated by this intuition, we propose a Dual Projection GAN (P2GAN) model that learns to balance between data matching and label matching. We then propose an improved cGAN model with Auxiliary Classification that directly aligns the fake and real conditionals P (class|image) by minimizing their f-divergence. Experiments on a synthetic Mixture of Gaussian (MoG) dataset and a variety of real-world datasets including CIFAR100, ImageNet, and VGGFace2 demonstrate the efficacy of our proposed models.

Recent studies have demonstrated the vulnerability of deep neural networks against adversarial examples. In-spired by the observation that adversarial examples often lie outside the natural image data manifold and the intrinsic dimension of image data is much smaller than its pixel space dimension, we propose to embed high-dimensional input images into a low-dimensional space and apply regularization on the embedding space to push the adversarial examples back to the manifold. The proposed framework is called Embedding Regularized Classifier (ER-Classifier), which improves the adversarial robustness of the classifier through embedding regularization. Besides improving classification accuracy against adversarial examples, the framework can be combined with detection methods to detect adversarial examples. Experimental results on several benchmark datasets show that, our proposed framework achieves good performance against strong adversarial at-tack methods.

MOTIVATION: Mapping distal regulatory elements, such as enhancers, is a cornerstone for elucidating how genetic variations may influence diseases. Previous enhancer-prediction methods have used either unsupervised approaches or supervised methods with limited training data. Moreover, past approaches have implemented enhancer discovery as a binary classification problem without accurate boundary detection, producing low-resolution annotations with superfluous regions and reducing the statistical power for downstream analyses (e.g. causal variant mapping and functional validations). Here, we addressed these challenges via a two-step model called Deep-learning framework for Condensing enhancers and refining boundaries with large-scale functional assays (DECODE). First, we employed direct enhancer-activity readouts from novel functional characterization assays, such as STARR-seq, to train a deep neural network for accurate cell-type-specific enhancer prediction. Second, to improve the annotation resolution, we implemented a weakly supervised object detection framework for enhancer localization with precise boundary detection (to a 10 bp resolution) using Gradient-weighted Class Activation Mapping. RESULTS: Our DECODE binary classifier outperformed a state-of-the-art enhancer prediction method by 24% in transgenic mouse validation. Furthermore, the object detection framework can condense enhancer annotations to only 13% of their original size, and these compact annotations have significantly higher conservation scores and genome-wide association study variant enrichments than the original predictions. Overall, DECODE is an effective tool for enhancer classification and precise localization. AVAILABILITY AND IMPLEMENTATION: DECODE source code and pre-processing scripts are available at decode.gersteinlab.org. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Pengyu Cheng, Martin Renqiang Min, Dinghan Shen, Christopher Malon, Yizhe Zhang, Yitong Li, Lawrence Carin. Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics. 2020.

We propose a sequential variational autoencoder to learn disentangled representations of sequential data (e.g., videos and audios) under self-supervision. Specifically, we exploit the benefits of some readily accessible supervision signals from input data itself or some off-the-shelf functional models and accordingly design auxiliary tasks for our model to utilize these signals. With the supervision of the signals, our model can easily disentangle the representation of an input sequence into static factors and dynamic factors (i.e., time-invariant and time-varying parts). Comprehensive experiments across videos and audios verify the effectiveness of our model on representation disentanglement and generation of sequential data, and demonstrate that, our model with self-supervision performs comparable to, if not better than, the fully-supervised model with ground truth labels, and outperforms state-of-the-art unsupervised models by a large margin.

The rich and accessible labeled data fueled the revolutionary successes of deep learning in object recognition. However, recognizing objects of novel classes with limited supervision information provided, i.e., Novel Object Recognition (NOR), remains a challenging task. We identify in this paper two key factors for the success of NOR that previous approaches fail to simultaneously guarantee. The first is producing discriminative feature representations for images of novel classes, and the second is generating a flexible classifier readily adapted to novel classes provided with limited supervision signals. To secure both key factors, we propose a framework which decouples a deep classification model into a feature extraction module and a classification module. We learn the former to ensure feature discriminability with a standard multi-class classification task by fully utilizing the competing information among all classes within a training set, and learn the latter to secure adaptability by training a meta-learner network which generates classifier weights whenever provided with minimal supervision information of target classes. Extensive experiments on common benchmark datasets in the settings of both zero-shot and few-shot learning demonstrate our method achieves state-of-the-art performance.

Zero-shot learning (ZSL) aims to recognize instances of unseen classes solely based on the semantic descriptions of the classes. Existing algorithms usually formulate it as a semantic-visual correspondence problem, by learning mappings from one feature space to the other. Despite being reasonable, previous approaches essentially discard the highly precious discriminative power of visual features in an implicit way, and thus produce undesirable results. We instead reformulate ZSL as a conditioned visual classification problem, i.e., classifying visual features based on the classifiers learned from the semantic descriptions. With this reformulation, we develop algorithms targeting various ZSL settings: For the conventional setting, we propose to train a deep neural network that directly generates visual feature classifiers from the semantic attributes with an episode-based training scheme; For the generalized setting, we concatenate the learned highly discriminative classifiers for seen classes and the generated classifiers for unseen classes to classify visual features of all classes; For the transductive setting, we exploit unlabeled data to effectively calibrate the classifier generator using a novel learning-without-forgetting self-training mechanism and guide the process by a robust generalized cross-entropy loss. Extensive experiments show that our proposed algorithms significantly outperform state-of-the-art methods by large margins on most benchmark datasets in all the ZSL settings.

Developing conditional generative models for text-to-video synthesis is an extremely challenging yet an important topic of research in machine learning. In this work, we address this problem by introducing Text-Filter conditioning Generative Adversarial Network (TFGAN), a conditional GAN model with a novel multi-scale text-conditioning scheme that improves text-video associations. By combining the proposed conditioning scheme with a deep GAN architecture, TFGAN generates high quality videos from text on challenging real-world video datasets. In addition, we construct a synthetic dataset of text-conditioned moving shapes to systematically evaluate our conditioning scheme. Extensive experiments demonstrate that TFGAN significantly outperforms existing approaches, and can also generate videos of novel categories not seen during training.

Traditional embedding approaches associate a real-valued embedding vector with each symbol or data point, which is equivalent to applying a linear transformation to ``one-hot" encoding of discrete symbols or data objects. Despite simplicity, these methods generate storage-inefficient representations and fail to effectively encode the internal semantic structure of data, especially when the number of symbols or data points and the dimensionality of the real-valued embedding vectors are large. In this paper, we propose a regularized autoencoder framework to learn compact Hierarchical K-way D-dimensional (HKD) discrete embedding of symbols or data points, aiming at capturing essential semantic structures of data. Experimental results on synthetic and real-world datasets show that our proposed HKD embedding can effectively reveal the semantic structure of data via hierarchical data visualization and greatly reduce the search space of nearest neighbor retrieval while preserving high accuracy.

In spite of its importance, passenger demand prediction is a highly challenging problem, because the demand is simultaneously influenced by the complex interactions among many spatial and temporal factors and other external factors such as weather. To address this problem, we propose a Spatio-TEmporal Fuzzy neural Network (STEF-Net) to accurately predict passenger demands incorporating the complex interactions of all known important factors. We design an end-to-end learning framework with different neural networks modeling different factors. Specifically, we propose to capture spatio-temporal feature interactions via a convolutional long short-term memory network and model external factors via a fuzzy neural network that handles data uncertainty significantly better than deterministic methods. To keep the temporal relations when fusing two networks and emphasize discriminative spatio-temporal feature interactions, we employ a novel feature fusion method with a convolution operation and an attention layer. As far as we know, our work is the first to fuse a deep recurrent neural network and a fuzzy neural network to model complex spatial-temporal feature interactions with additional uncertain input features for predictive learning. Experiments on a large-scale real-world dataset show that our model achieves more than 10% improvement over the state-of-the-art approaches.

Generating videos from text has proven to be a significant challenge for existing generative models. We tackle this problem by training a conditional generative model to extract both static and dynamic information from text. This is manifested in a hybrid framework, employing a Variational Autoencoder (VAE) and a Generative Adversarial Network (GAN). The static features, called "gist," are used to sketch text-conditioned background color and object layout structure. Dynamic features are considered by transforming input text into an image filter. To obtain a large amount of data for training the deep-learning model, we develop a method to automatically create a matched text-video corpus from publicly available online videos. Experimental results show that the proposed framework generates plausible and diverse short-duration smooth videos, while accurately reflecting the input text information. It significantly outperforms baseline models that directly adapt text-to-image generation procedures to produce videos. Performance is evaluated both visually and by adapting the inception score used to evaluate image generation in GANs.

Previous models for video captioning often use the output from a specific layer of a Convolutional Neural Network (CNN) as video features. However, the variable context-dependent semantics in the video may make it more appropriate to adaptively select features from the multiple CNN layers. We propose a new approach for generating adaptive spatiotemporal representations of videos for the captioning task. A novel attention mechanism is developed, that adaptively and sequentially focuses on different layers of CNN features (levels of feature "abstraction"), as well as local spatiotemporal regions of the feature maps at each layer. The proposed approach is evaluated on three benchmark datasets: YouTube2Text, M-VAD and MSR-VTT. Along with visualizing the results and how the model works, these experiments quantitatively demonstrate the effectiveness of the proposed adaptive spatiotemporal feature abstraction for translating videos to sentences with rich semantics.

Dinghan Shen, Guoyin Wang, Wenlin Wang, Martin Renqiang Min, Qinliang Su, Yizhe Zhang, Chunyuan Li, Ricardo Henao, Lawrence Carin. Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2018.

Convolutional neural networks (CNNs) have recently emerged as a popular building block for natural language processing (NLP).Despite their success, most existing CNN models employed in NLP share the same learned (and static) set of filters for all input sentences.In this paper, we consider an approach of using a small meta network to learn contextaware convolutional filters for text processing.The role of meta network is to abstract the contextual information of a sentence or document into a set of input-aware filters.We further generalize this framework to model sentence pairs, where a bidirectional filter generation mechanism is introduced to encapsulate co-dependent sentence representations.In our benchmarks on four different tasks, including ontology classification, sentiment analysis, answer sentence selection, and paraphrase identification, our proposed model, a modified CNN with context-aware filters, consistently outperforms the standard CNN and attentionbased CNN baselines.By visualizing the learned context-aware filters, we further validate and rationalize the effectiveness of proposed framework.

Conventional embedding methods directly associate each symbol with a continuous embedding vector, which is equivalent to applying a linear transformation based on a “one-hot” encoding of the discrete symbols. Despite its simplicity, such approach yields the number of parameters that grows linearly with the vocabulary size and can lead to overfitting. In this work, we propose a much more compact K-way D-dimensional discrete encoding scheme to replace the “one-hot" encoding. In the proposed “KD encoding”, each symbol is represented by a $D$-dimensional code with a cardinality of $K$, and the final symbol embedding vector is generated by composing the code embedding vectors. To end-to-end learn semantically meaningful codes, we derive a relaxed discrete optimization approach based on stochastic gradient descent, which can be generally applied to any differentiable computational graph with an embedding layer. In our experiments with various applications from natural language processing to graph convolutional networks, the total size of the embedding layer can be reduced up to 98% while achieving similar or better performance.

Metric learning methods for dimensionality reduction in combination with k-Nearest Neighbors (kNN) have been extensively deployed in many classification, data embedding, and information retrieval applications. However, most of these approaches involve pairwise training data comparisons, and thus have quadratic computational complexity with respect to the size of training set, preventing them from scaling to fairly big datasets. Moreover, during testing, comparing test data against all the training data points is also expensive in terms of both computational cost and resources required. Furthermore, previous metrics are either too constrained or too expressive to be well learned. To effectively solve these issues, we present an exemplar-centered supervised shallow parametric data embedding model, using a Maximally Collapsing Metric Learning (MCML) objective. Our strategy learns a shallow high-order parametric embedding function and compares training/test data only with learned or precomputed exemplars, resulting in a cost function with linear computational complexity for both training and testing. We also empirically demonstrate, using several benchmark datasets, that for classification in two-dimensional embedding space, our approach not only gains speedup of kNN by hundreds of times, but also outperforms state-of-the-art supervised embedding approaches.

Neural network based sequence-to-sequence models in an encoder-decoder framework have been successfully applied to solve Question Answering (QA) problems, predicting answers from statements and questions. However, almost all previous models have failed to consider detailed context information and unknown states under which systems do not have enough information to answer given questions. These scenarios with incomplete or ambiguous information are very common in the setting of Interactive Question Answering (IQA). To address this challenge, we develop a novel model, employing context-dependent word-level attention for more accurate statement representations and question-guided sentence-level attention for better context modeling. We also generate unique IQA datasets to test our model, which will be made publicly available. Employing these attention mechanisms, our model accurately understands when it can output an answer or when it requires generating a supplementary question for additional input depending on different contexts. When available, user's feedback is encoded and directly applied to update sentence-level attention to infer an answer. Extensive experiments on QA and IQA datasets quantitatively demonstrate the effectiveness of our model with significant improvement over state-of-the-art conventional QA models.

In mission critical IT services, system failure prediction becomes increasingly important; it prevents unexpected system downtime, and assures service reliability for end users. While operational console logs record rich and descriptive information on the health status of those IT systems, existing system management technologies mostly use them in a labor-intensive forensics approach, i.e., identifying what went wrong after the fact. Recent efforts on log-based system management take an automation approach with text mining techniques, such as term frequency - inverse document frequency (TF-IDF). However, those techniques lead to a high-dimensional feature space, and are not easily generalizable to heterogeneous log formats. In this paper, we present a novel system that automatically parses streamed console logs and detects early warning signals for IT system failure prediction. In particular, our solution includes a log pattern extraction method by clustering together logs with similar format and content. We then resemble the TF-IDF idea by considering each pattern as a word and the set of patterns in each discretized epoch as a document. This leads to a feature space with significantly lower dimensionality that can provide robust signals for the status of the system. As system failures tend to occur very rare, we apply a recurrent neural network, namely, Long Short-Term Memory (LSTM), to deal with the “rarity” of labeled data in the training process. LSTM is able to capture the long-range dependency across sequences, therefore outperforms traditional supervised learning methods in our application domain. We evaluated and compared our proposed technology with state-of-the-art machine learning approaches using real log traces from two large enterprise systems. The results showed the advantage and potentials of our system in prediction of complex IT failures. To our knowledge, our work is the first that employs LSTM for log-based system failure prediction.

MOTIVATION: Effective computational methods for peptide-protein binding prediction can greatly help clinical peptide vaccine search and design. However, previous computational methods fail to capture key nonlinear high-order dependencies between different amino acid positions. As a result, they often produce low-quality rankings of strong binding peptides. To solve this problem, we propose nonlinear high-order machine learning methods including high-order neural networks (HONNs) with possible deep extensions and high-order kernel support vector machines to predict major histocompatibility complex-peptide binding. RESULTS: The proposed high-order methods improve quality of binding predictions over other prediction methods. With the proposed methods, a significant gain of up to 25-40% is observed on the benchmark and reference peptide datasets and tasks. In addition, for the first time, our experiments show that pre-training with high-order semi-restricted Boltzmann machines significantly improves the performance of feed-forward HONNs. Moreover, our experiments show that the proposed shallow HONN outperform the popular pre-trained deep neural network on most tasks, which demonstrates the effectiveness of modelling high-order feature interactions for predicting major histocompatibility complex-peptide binding. AVAILABILITY AND IMPLEMENTATION: There is no associated distributable software. CONTACT: [email protected] or [email protected] SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Fully-observable high-order Boltzmann Machines are capable of identifying explicit high-order feature interactions theoretically. However, they have never been used in practice due to their prohibitively high computational cost for inference and learning. In this paper, we propose an efficient approach for learning a fully observable high-order Boltzmann Machine based on sparse learning and contrastive divergence, resulting in an interpretable Sparse High-order Boltzmann Machine, denoted as SHBM. Experimental results on synthetic datasets and a real dataset demonstrate that SHBM can produce higher pseudo-log-likelihood and better reconstructions on test data than the state-of-the-art methods. In addition, we apply SHBM to a challenging bioinformatics problem of discovering complex Transcription Factor interactions. Compared to conventional Boltzmann Machine and directed Bayesian Network, SHBM can identify much more biologically meaningful interactions that are supported by recent biological studies. To the best of our knowledge, SHBM is the first working Boltzmann Machine with explicit high-order feature interactions applied to real-world problems.

Identifying interpretable discriminative high-order feature interactions given limited training data in high dimensions is challenging in both machine learning and data mining. In this paper, we propose a factorization based sparse learning framework termed FHIM for identifying high-order feature interactions in linear and logistic regression models, and study several optimization methods for solving them. Unlike previous sparse learning methods, our model FHIM recovers both the main effects and the interaction terms accurately without imposing tree-structured hierarchical constraints. Furthermore, we show that FHIM has oracle properties when extended to generalized linear regression models with pairwise interactions. Experiments on simulated data show that FHIM outperforms the state-of-the-art sparse lear-ning techniques. Further experiments on our experimentally generated data from patient blood samples using a novel SOMAmer (Slow Off-rate Modified Aptamer) technology show that, FHIM performs blood-based cancer diagnosis and bio-marker discovery for Renal Cell Carcinoma much better than other competing methods, and it identifies interpretable block-wise high-order gene interactions predictive of cancer stages of samples. A literature survey shows that the interactions identified by FHIM play important roles in cancer development.

MOTIVATION: ChIP-seq and ChIP-chip experiments have been widely used to identify transcription factor (TF) binding sites and target genes. Conventionally, a fairly 'simple' approach is employed for target gene identification e.g. finding genes with binding sites within 2 kb of a transcription start site (TSS). However, this does not take into account the number of sites upstream of the TSS, their exact positioning or the fact that different TFs appear to act at different characteristic distances from the TSS. RESULTS: Here we propose a probabilistic model called target identification from profiles (TIP) that quantitatively measures the regulatory relationships between TFs and target genes. For each TF, our model builds a characteristic, averaged profile of binding around the TSS and then uses this to weight the sites associated with a given gene, providing a continuous-valued 'regulatory' score relating each TF and potential target. Moreover, the score can readily be turned into a ranked list of target genes and an estimate of significance, which is useful for case-dependent downstream analysis. CONCLUSION: We show the advantages of TIP by comparing it to the 'simple' approach on several representative datasets, using motif occurrence and relationship to knock-out experiments as metrics of validation. Moreover, we show that the probabilistic model is not as sensitive to various experimental parameters (including sequencing depth and peak-calling method) as the simple approach; in fact, the lesser dependence on sequencing depth potentially utilizes the result of a ChIP-seq experiment in a more 'cost-effective' manner. CONTACT: [email protected] SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Variations in gene expression level might lead to phenotypic diversity across individuals or populations. Although many human genes are found to have differential mRNA levels between populations, the extent of gene expression that could vary within and between populations largely remains elusive. To investigate the dynamic range of gene expression, we analyzed the expression variability of ∼18, 000 human genes across individuals within HapMap populations. Although ∼20% of human genes show differentiated mRNA levels between populations, our results show that expression variability of most human genes in one population is not significantly deviant from another population, except for a small fraction that do show substantially higher expression variability in a particular population. By associating expression variability with sequence polymorphism, intriguingly, we found SNPs in the untranslated regions (5' and 3'UTRs) of these variable genes show consistently elevated population heterozygosity. We performed differential expression analysis on a genome-wide scale, and found substantially reduced expression variability for a large number of genes, prohibiting them from being differentially expressed between populations. Functional analysis revealed that genes with the greatest within-population expression variability are significantly enriched for chemokine signaling in HIV-1 infection, and for HIV-interacting proteins that control viral entry, replication, and propagation. This observation combined with the finding that known human HIV host factors show substantially elevated expression variability, collectively suggest that gene expression variability might explain differential HIV susceptibility across individuals.

KNN is one of the most popular data mining methods for classification, but it often fails to work well with inappropriate choice of distance metric or due to the presence of numerous class-irrelevant features. Linear feature transformation methods have been widely applied to extract class-relevant information to improve kNN classification, which is very limited in many applications. Kernels have also been used to learn powerful non-linear feature transformations, but these methods fail to scale to large datasets. In this paper, we present a scalable non-linear feature mapping method based on a deep neural network pretrained with Restricted Boltzmann Machines for improving kNN classification in a large-margin framework, which we call DNet-kNN. DNet-kNN can be used for both classification and for supervised dimensionality reduction. The experimental results on two benchmark handwritten digit datasets and one newsgroup text dataset show that DNet-kNN has much better performance than large-margin kNN using a linear mapping and kNN based on a deep autoencoder pretrained with Restricted Boltzmann Machines.

Random-walk based algorithms are good choices for solving many classification problems with limited labeled data and a large amount of unlabeled data. However, it is difficult to choose the optimal number of random steps, and the results are very sensitive to the parameter chosen. In this paper, we will discuss how to better identify protein remote homology than any other algorithm using a learned random-walk kernel based on a positive linear combination of random-walk kernels with different random steps, which leads to a convex combination of kernels. The resulting kernel has much better prediction performance than the state-of-the-art profile kernel for protein remote homology identification. On the SCOP benchmark dataset, the overall mean ROC50 score on 54 protein families we obtained using the new kernel is above 0.90, which has almost perfect prediction performance on most of the 54 families and has significant improvement over the best published result; moreover, our approach based on learned random-walk kernels can effectively identify meaningful protein sequence motifs that are responsible for discriminating the memberships of protein sequences' remote homology in SCOP.

Kernel learning methods based on kernel alignment with semidefinite programming (SDP) are often memory intensive and computationally expensive, thus often impractical for problems with large-size dataset. We propose a method using label information to modify kernels based on SVD and a linear mapping. As a result, the new kernel matrix reflects the label-dependent separability of the data in a better way than the original kernel matrix. In addition, our experimental results on USPS handwritten digits and the SCOP dataset, show that the SVM classifier based on the improved kernels has better performance than the SVM classifier based on the original kernels; moreover, SVM based on the improved profile kernel with pull-in homologs (see experiment section for explanations) produced the best results for remote homology detection on the SCOP dataset compared to the published results.