Sheng Gao

Jia Li, Ziyang Zhang, Junjie Lang et al.

In this paper, we present our solutions for the Multimodal Sentiment Analysis Challenge (MuSe) 2022, which includes MuSe-Humor, MuSe-Reaction and MuSe-Stress Sub-challenges. The MuSe 2022 focuses on humor detection, emotional reactions and multimodal emotional stress utilizing different modalities and data sets. In our work, different kinds of multimodal features are extracted, including acoustic, visual, text and biological features. These features are fused by TEMMA and GRU with self-attention mechanism frameworks. In this paper, 1) several new audio features, facial expression features and paragraph-level text embeddings are extracted for accuracy improvement. 2) we substantially improve the accuracy and reliability of multimodal sentiment prediction by mining and blending the multimodal features. 3) effective data augmentation strategies are applied in model training to alleviate the problem of sample imbalance and prevent the model from learning biased subject characters. For the MuSe-Humor sub-challenge, our model obtains the AUC score of 0.8932. For the MuSe-Reaction sub-challenge, the Pearson's Correlations Coefficient of our approach on the test set is 0.3879, which outperforms all other participants. For the MuSe-Stress sub-challenge, our approach outperforms the baseline in both arousal and valence on the test dataset, reaching a final combined result of 0.5151.

Ziyang Zheng, Zhengyang Duan, Hang Chen et al.

Photonic neural network (PNN) is a remarkable analog artificial intelligence (AI) accelerator that computes with photons instead of electrons to feature low latency, high energy efficiency, and high parallelism. However, the existing training approaches cannot address the extensive accumulation of systematic errors in large-scale PNNs, resulting in a significant decrease in model performance in physical systems. Here, we propose dual adaptive training (DAT) that allows the PNN model to adapt to substantial systematic errors and preserves its performance during the deployment. By introducing the systematic error prediction networks with task-similarity joint optimization, DAT achieves the high similarity mapping between the PNN numerical models and physical systems and high-accurate gradient calculations during the dual backpropagation training. We validated the effectiveness of DAT by using diffractive PNNs and interference-based PNNs on image classification tasks. DAT successfully trained large-scale PNNs under major systematic errors and preserved the model classification accuracies comparable to error-free systems. The results further demonstrated its superior performance over the state-of-the-art in situ training approaches. DAT provides critical support for constructing large-scale PNNs to achieve advanced architectures and can be generalized to other types of AI systems with analog computing errors.

Hua Wang, Sheng Gao, Huanyu Zhang et al.

Many modern machine learning algorithms are composed of simple private algorithms; thus, an increasingly important problem is to efficiently compute the overall privacy loss under composition. In this study, we introduce the Edgeworth Accountant, an analytical approach to composing differential privacy guarantees of private algorithms. The Edgeworth Accountant starts by losslessly tracking the privacy loss under composition using the $f$-differential privacy framework, which allows us to express the privacy guarantees using privacy-loss log-likelihood ratios (PLLRs). As the name suggests, this accountant next uses the Edgeworth expansion to the upper and lower bounds the probability distribution of the sum of the PLLRs. Moreover, by relying on a technique for approximating complex distributions using simple ones, we demonstrate that the Edgeworth Accountant can be applied to the composition of any noise-addition mechanism. Owing to certain appealing features of the Edgeworth expansion, the $(ε, δ)$-differential privacy bounds offered by this accountant are non-asymptotic, with essentially no extra computational cost, as opposed to the prior approaches in, wherein the running times increase with the number of compositions. Finally, we demonstrate that our upper and lower $(ε, δ)$-differential privacy bounds are tight in federated analytics and certain regimes of training private deep learning models.

Hua Wang, Sheng Gao, Huanyu Zhang et al.

Hyperparameter optimization, also known as hyperparameter tuning, is a widely recognized technique for improving model performance. Regrettably, when training private ML models, many practitioners often overlook the privacy risks associated with hyperparameter optimization, which could potentially expose sensitive information about the underlying dataset. Currently, the sole existing approach to allow privacy-preserving hyperparameter optimization is to uniformly and randomly select hyperparameters for a number of runs, subsequently reporting the best-performing hyperparameter. In contrast, in non-private settings, practitioners commonly utilize ``adaptive'' hyperparameter optimization methods such as Gaussian process-based optimization, which select the next candidate based on information gathered from previous outputs. This substantial contrast between private and non-private hyperparameter optimization underscores a critical concern. In our paper, we introduce DP-HyPO, a pioneering framework for ``adaptive'' private hyperparameter optimization, aiming to bridge the gap between private and non-private hyperparameter optimization. To accomplish this, we provide a comprehensive differential privacy analysis of our framework. Furthermore, we empirically demonstrate the effectiveness of DP-HyPO on a diverse set of real-world datasets.

Yuchen Liu, Yingjie Feng, Lixiong Qin et al.

In Agentic Search, trajectory-level outcome rewards fail to quantify the behavioral contributions of individual steps, while existing step-level reward methods typically rely on costly tree sampling. We view world knowledge as a latent world graph and each IS task as search within a latent task graph, where effective steps should make graph progress toward the answer node. Based on this prior, we propose Graph-Distance Contribution Reward (GDCR), a step-level process reward that scores newly-retrieved and newly-cited entities by their distance to the answer node in a training-time Entity-Relation (ER) graph. We further propose Step Advantage Policy Optimization (SAPO), which converts GDCR into step-level advantages and combines them with trajectory-level outcome advantages. Experiments on four challenging benchmarks validate the effectiveness of our method.

Maria Cuellar, Sheng Gao, Heike Hofmann

Forensic toolmark analysis traditionally relies on subjective human judgment, leading to inconsistencies and lack of transparency. The multitude of variables, including angles and directions of mark generation, further complicates comparisons. To address this, we first generate a dataset of 3D toolmarks from various angles and directions using consecutively manufactured slotted screwdrivers. By using PAM clustering, we find that there is clustering by tool rather than angle or direction. Using Known Match and Known Non-Match densities, we establish thresholds for classification. Fitting Beta distributions to the densities, we allow for the derivation of likelihood ratios for new toolmark pairs. With a cross-validated sensitivity of 98% and specificity of 96%, our approach enhances the reliability of toolmark analysis. This approach is applicable to slotted screwdrivers, and for screwdrivers that are made with a similar production method. With data collection of other tools and factors, it could be applied to compare toolmarks of other types. This empirically trained, open-source solution offers forensic examiners a standardized means to objectively compare toolmarks, potentially decreasing the number of miscarriages of justice in the legal system.

Shanzhuo Zhang, Lihang Liu, Sheng Gao et al.

In this report, we (SuperHelix team) present our solution to KDD Cup 2021-PCQM4M-LSC, a large-scale quantum chemistry dataset on predicting HOMO-LUMO gap of molecules. Our solution, Lite Geometry Enhanced Molecular representation learning (LiteGEM) achieves a mean absolute error (MAE) of 0.1204 on the test set with the help of deep graph neural networks and various self-supervised learning tasks. The code of the framework can be found in https://github.com/PaddlePaddle/PaddleHelix/tree/dev/competition/kddcup2021-PCQM4M-LSC/.

Kun Liu, Yao Fu, Chuanqi Tan et al.

Recent studies in deep learning have shown significant progress in named entity recognition (NER). Most existing works assume clean data annotation, yet a fundamental challenge in real-world scenarios is the large amount of noise from a variety of sources (e.g., pseudo, weak, or distant annotations). This work studies NER under a noisy labeled setting with calibrated confidence estimation. Based on empirical observations of different training dynamics of noisy and clean labels, we propose strategies for estimating confidence scores based on local and global independence assumptions. We partially marginalize out labels of low confidence with a CRF model. We further propose a calibration method for confidence scores based on the structure of entity labels. We integrate our approach into a self-training framework for boosting performance. Experiments in general noisy settings with four languages and distantly labeled settings demonstrate the effectiveness of our method. Our code can be found at https://github.com/liukun95/Noisy-NER-Confidence-Estimation

Sheng Gao, Songtao Yang, Haiou Zhang et al.

The rapid progress in radar and communication places increasing demands on low-latency and energy-efficiency array signal processing methods. There is an emerging direction of constructing analog computing processors for directly processing electromagnetic (EM) waves. However, the existing methods are constrained by 2D physical aperture and imprecise design process with inefficient computing architecture, resulting in limited sensing resolution and number of separated sources. Here, we present a fully-analog array signal processor (FASP) using 3D aperture engineering framework to perform super-resolution direction-of-arrival estimation, source number estimation, and multi-channel source separation in parallel for both coherent and incoherent sources. 3D aperture engineering is realized by constructing deep cascaded metasurface layers so that the diffractive propagation from oblique incident fields can be layer-wise modulated and piecewise encoded for perceiving EM fields far exceeding physical aperture limits. The multi-dimensional synthetic aperture (MSA) training is developed to characterize the metasurface modulation and optimize the neuro-augmented physical model for extending system aperture and generating high-order nonlinear angular response. FASP orthogonalizes the array response vectors of communication channels to map them into antenna detectors in the analog domain. The $N$-layer FASP has the capability to achieve ~N times higher angular resolution than the Rayleigh diffraction limit. Experiments further validate the source number estimation and independent channel separation of 10-target that can suppress radar jamming signals by ~20 dB and enhance channel communication capacity by 13.5 times at 36~41 GHz. FASP heralds a paradigm shift in signal processing for super-resolution optics, advanced radar, and 6G communications.

Songtao Yang, Sheng Gao, Chu Wu et al.

Diffractive neural networks leverage the high-dimensional characteristics of electromagnetic (EM) fields for high-throughput computing. However, the existing architectures face challenges in integrating large-scale multidimensional metasurfaces with precise network training and haven't utilized multidimensional EM field coding scheme for super-resolution sensing. Here, we propose diffractive meta-neural networks (DMNNs) for accurate EM field modulation through metasurfaces, which enable multidimensional multiplexing and coding for multi-task learning and high-throughput super-resolution direction of arrival estimation. DMNN integrates pre-trained mini-metanets to characterize the amplitude and phase responses of meta-atoms across different polarizations and frequencies, with structure parameters inversely designed using the gradient-based meta-training. For wide-field super-resolution angle estimation, the system simultaneously resolves azimuthal and elevational angles through x and y-polarization channels, while the interleaving of frequency-multiplexed angular intervals generates spectral-encoded optical super-oscillations to achieve full-angle high-resolution estimation. Post-processing lightweight electronic neural networks further enhance the performance. Experimental results validate that a three-layer DMNN operating at 27 GHz, 29 GHz, and 31 GHz achieves $\sim7\times$ Rayleigh diffraction-limited angular resolution (0.5$^\circ$), a mean absolute error of 0.048$^\circ$ for two incoherent targets within a $\pm 11.5^\circ$ field of view, and an angular estimation throughput an order of magnitude higher (1917) than that of existing methods. The proposed architecture advances high-dimensional photonic computing systems by utilizing inherent high-parallelism and all-optical coding methods for ultra-high-resolution, high-throughput applications.

Hang Chen, Sheng Gao, Zejia Zhao et al.

Optical super-oscillation enables far-field super-resolution imaging beyond diffraction limits. However, the existing super-oscillatory lens for the spatial super-resolution imaging system still confronts critical limitations in performance due to the lack of a more advanced design method and the limited design degree of freedom. Here, we propose an optical super-oscillatory diffractive neural network, i.e., SODNN, that can achieve super-resolved spatial resolution for imaging beyond the diffraction limit with superior performance over existing methods. SODNN is constructed by utilizing diffractive layers to implement optical interconnections and imaging samples or biological sensors to implement nonlinearity, which modulates the incident optical field to create optical super-oscillation effects in 3D space and generate the super-resolved focal spots. By optimizing diffractive layers with 3D optical field constraints under an incident wavelength size of $λ$, we achieved a super-oscillatory spot with a full width at half maximum of 0.407$λ$ in the far field distance over 400$λ$ without side-lobes over the field of view, having a long depth of field over 10$λ$. Furthermore, the SODNN implements a multi-wavelength and multi-focus spot array that effectively avoids chromatic aberrations. Our research work will inspire the development of intelligent optical instruments to facilitate the applications of imaging, sensing, perception, etc.

Sheng Gao, Zongming Ma

Generalized correlation analysis (GCA) is concerned with uncovering linear relationships across multiple datasets. It generalizes canonical correlation analysis that is designed for two datasets. We study sparse GCA when there are potentially multiple generalized correlation tuples in data and the loading matrix has a small number of nonzero rows. It includes sparse CCA and sparse PCA of correlation matrices as special cases. We first formulate sparse GCA as generalized eigenvalue problems at both population and sample levels via a careful choice of normalization constraints. Based on a Lagrangian form of the sample optimization problem, we propose a thresholded gradient descent algorithm for estimating GCA loading vectors and matrices in high dimensions. We derive tight estimation error bounds for estimators generated by the algorithm with proper initialization. We also demonstrate the prowess of the algorithm on a number of synthetic datasets.

Jingxuan Yang, Kerui Xu, Jun Xu et al.

In this paper, we present a neural model for joint dropped pronoun recovery (DPR) and conversational discourse parsing (CDP) in Chinese conversational speech. We show that DPR and CDP are closely related, and a joint model benefits both tasks. We refer to our model as DiscProReco, and it first encodes the tokens in each utterance in a conversation with a directed Graph Convolutional Network (GCN). The token states for an utterance are then aggregated to produce a single state for each utterance. The utterance states are then fed into a biaffine classifier to construct a conversational discourse graph. A second (multi-relational) GCN is then applied to the utterance states to produce a discourse relation-augmented representation for the utterances, which are then fused together with token states in each utterance as input to a dropped pronoun recovery layer. The joint model is trained and evaluated on a new Structure Parsing-enhanced Dropped Pronoun Recovery (SPDPR) dataset that we annotated with both two types of information. Experimental results on the SPDPR dataset and other benchmarks show that DiscProReco significantly outperforms the state-of-the-art baselines of both tasks.

Jingxuan Yang, Kerui Xu, Jun Xu et al.

Pronouns are often dropped in Chinese conversations and recovering the dropped pronouns is important for NLP applications such as Machine Translation. Existing approaches usually formulate this as a sequence labeling task of predicting whether there is a dropped pronoun before each token and its type. Each utterance is considered to be a sequence and labeled independently. Although these approaches have shown promise, labeling each utterance independently ignores the dependencies between pronouns in neighboring utterances. Modeling these dependencies is critical to improving the performance of dropped pronoun recovery. In this paper, we present a novel framework that combines the strength of Transformer network with General Conditional Random Fields (GCRF) to model the dependencies between pronouns in neighboring utterances. Results on three Chinese conversation datasets show that the Transformer-GCRF model outperforms the state-of-the-art dropped pronoun recovery models. Exploratory analysis also demonstrates that the GCRF did help to capture the dependencies between pronouns in neighboring utterances, thus contributes to the performance improvements.

Jingxuan Yang, Jianzhuo Tong, Si Li et al.

Pronouns are often dropped in Chinese sentences, and this happens more frequently in conversational genres as their referents can be easily understood from context. Recovering dropped pronouns is essential to applications such as Information Extraction where the referents of these dropped pronouns need to be resolved, or Machine Translation when Chinese is the source language. In this work, we present a novel end-to-end neural network model to recover dropped pronouns in conversational data. Our model is based on a structured attention mechanism that models the referents of dropped pronouns utilizing both sentence-level and word-level information. Results on three different conversational genres show that our approach achieves a significant improvement over the current state of the art.

Jingxuan Yang, Jun Xu, Jianzhuo Tong et al.

Next basket recommendation, which aims to predict the next a few items that a user most probably purchases given his historical transactions, plays a vital role in market basket analysis. From the viewpoint of item, an item could be purchased by different users together with different items, for different reasons. Therefore, an ideal recommender system should represent an item considering its transaction contexts. Existing state-of-the-art deep learning methods usually adopt the static item representations, which are invariant among all of the transactions and thus cannot achieve the full potentials of deep learning. Inspired by the pre-trained representations of BERT in natural language processing, we propose to conduct context-aware item representation for next basket recommendation, called Item Encoder Representations from Transformers (IERT). In the offline phase, IERT pre-trains deep item representations conditioning on their transaction contexts. In the online recommendation phase, the pre-trained model is further fine-tuned with an additional output layer. The output contextualized item embeddings are used to capture users' sequential behaviors and general tastes to conduct recommendation. Experimental results on the Ta-Feng data set show that IERT outperforms the state-of-the-art baseline methods, which demonstrated the effectiveness of IERT in next basket representation.

Hengru Xu, Shen Li, Renfen Hu et al.

Dropout is used to avoid overfitting by randomly dropping units from the neural networks during training. Inspired by dropout, this paper presents GI-Dropout, a novel dropout method integrating with global information to improve neural networks for text classification. Unlike the traditional dropout method in which the units are dropped randomly according to the same probability, we aim to use explicit instructions based on global information of the dataset to guide the training process. With GI-Dropout, the model is supposed to pay more attention to inapparent features or patterns. Experiments demonstrate the effectiveness of the dropout with global information on seven text classification tasks, including sentiment analysis and topic classification.

Yadi Lao, Jun Xu, Yanyan Lan et al.

In this paper we propose a novel reinforcement learning based model for sequence tagging, referred to as MM-Tag. Inspired by the success and methodology of the AlphaGo Zero, MM-Tag formalizes the problem of sequence tagging with a Monte Carlo tree search (MCTS) enhanced Markov decision process (MDP) model, in which the time steps correspond to the positions of words in a sentence from left to right, and each action corresponds to assign a tag to a word. Two long short-term memory networks (LSTM) are used to summarize the past tag assignments and words in the sentence. Based on the outputs of LSTMs, the policy for guiding the tag assignment and the value for predicting the whole tagging accuracy of the whole sentence are produced. The policy and value are then strengthened with MCTS, which takes the produced raw policy and value as inputs, simulates and evaluates the possible tag assignments at the subsequent positions, and outputs a better search policy for assigning tags. A reinforcement learning algorithm is proposed to train the model parameters. Our work is the first to apply the MCTS enhanced MDP model to the sequence tagging task. We show that MM-Tag can accurately predict the tags thanks to the exploratory decision making mechanism introduced by MCTS. Experimental results show based on a chunking benchmark showed that MM-Tag outperformed the state-of-the-art sequence tagging baselines including CRF and CRF with LSTM.

Siting Ren, Sheng Gao

Cross-domain recommendation has been proposed to transfer user behavior pattern by pooling together the rating data from multiple domains to alleviate the sparsity problem appearing in single rating domains. However, previous models only assume that multiple domains share a latent common rating pattern based on the user-item co-clustering. To capture diversities among different domains, we propose a novel Probabilistic Cluster-level Latent Factor (PCLF) model to improve the cross-domain recommendation performance. Experiments on several real world datasets demonstrate that our proposed model outperforms the state-of-the-art methods for the cross-domain recommendation task.

Sheng Gao, Ludovic Denoyer, Patrick Gallinari

This paper aims at the problem of link pattern prediction in collections of objects connected by multiple relation types, where each type may play a distinct role. While common link analysis models are limited to single-type link prediction, we attempt here to capture the correlations among different relation types and reveal the impact of various relation types on performance quality. For that, we define the overall relations between object pairs as a \textit{link pattern} which consists in interaction pattern and connection structure in the network, and then use tensor formalization to jointly model and predict the link patterns, which we refer to as \textit{Link Pattern Prediction} (LPP) problem. To address the issue, we propose a Probabilistic Latent Tensor Factorization (PLTF) model by introducing another latent factor for multiple relation types and furnish the Hierarchical Bayesian treatment of the proposed probabilistic model to avoid overfitting for solving the LPP problem. To learn the proposed model we develop an efficient Markov Chain Monte Carlo sampling method. Extensive experiments are conducted on several real world datasets and demonstrate significant improvements over several existing state-of-the-art methods.

Sheng Gao, Ludovic Denoyer, Patrick Gallinari

In this paper we address the problem of modeling relational data, which appear in many applications such as social network analysis, recommender systems and bioinformatics. Previous studies either consider latent feature based models but disregarding local structure in the network, or focus exclusively on capturing local structure of objects based on latent blockmodels without coupling with latent characteristics of objects. To combine the benefits of the previous work, we propose a novel model that can simultaneously incorporate the effect of latent features and covariates if any, as well as the effect of latent structure that may exist in the data. To achieve this, we model the relation graph as a function of both latent feature factors and latent cluster memberships of objects to collectively discover globally predictive intrinsic properties of objects and capture latent block structure in the network to improve prediction performance. We also develop an optimization transfer algorithm based on the generalized EM-style strategy to learn the latent factors. We prove the efficacy of our proposed model through the link prediction task and cluster analysis task, and extensive experiments on the synthetic data and several real world datasets suggest that our proposed LFBM model outperforms the other state of the art approaches in the evaluated tasks.