Hao Liu

Hao Wang, Junchao Liao, Tianheng Cheng et al.

Recently, the semantics of scene text has been proven to be essential in fine-grained image classification. However, the existing methods mainly exploit the literal meaning of scene text for fine-grained recognition, which might be irrelevant when it is not significantly related to objects/scenes. We propose an end-to-end trainable network that mines implicit contextual knowledge behind scene text image and enhance the semantics and correlation to fine-tune the image representation. Unlike the existing methods, our model integrates three modalities: visual feature extraction, text semantics extraction, and correlating background knowledge to fine-grained image classification. Specifically, we employ KnowBert to retrieve relevant knowledge for semantic representation and combine it with image features for fine-grained classification. Experiments on two benchmark datasets, Con-Text, and Drink Bottle, show that our method outperforms the state-of-the-art by 3.72\% mAP and 5.39\% mAP, respectively. To further validate the effectiveness of the proposed method, we create a new dataset on crowd activity recognition for the evaluation. The source code and new dataset of this work are available at https://github.com/lanfeng4659/KnowledgeMiningWithSceneText.

Hao Liu, Carmelo Sferrazza, Pieter Abbeel

Learning from human preferences is important for language models to match human needs and to align with human and social values. Prior works have achieved remarkable successes by learning from human feedback to understand and follow instructions. Nonetheless, these methods are either founded on hand-picked model generations that are favored by human annotators, rendering them inefficient in terms of data utilization and challenging to apply in general, or they depend on reinforcement learning, which often suffers from imperfect reward functions and relies on extremely challenging optimizations. In this work, we propose a novel technique, Chain of Hindsight, that is easy to optimize and can learn from any form of feedback, regardless of its polarity. Our idea is inspired by how humans learn from extensive feedback presented in the form of languages. We convert all types of feedback into sequences of sentences, which are then used to fine-tune the model, allowing us to take advantage of the language comprehension capabilities of language models. We condition the model on a sequence of model generations paired with feedback. By doing so, the model is trained to generate outputs based on feedback, while learning to identify and correct negative attributes or errors. Applying our method to large language models, we observed that Chain of Hindsight significantly surpasses previous methods in aligning language models with human preferences. We report significant improvements on summarization and dialogue benchmarks, with our approach markedly preferred in human evaluations.

Yasser Abduallah, Vania K. Jordanova, Hao Liu et al.

Solar energetic particles (SEPs) are an essential source of space radiation, which are hazards for humans in space, spacecraft, and technology in general. In this paper we propose a deep learning method, specifically a bidirectional long short-term memory (biLSTM) network, to predict if an active region (AR) would produce an SEP event given that (i) the AR will produce an M- or X-class flare and a coronal mass ejection (CME) associated with the flare, or (ii) the AR will produce an M- or X-class flare regardless of whether or not the flare is associated with a CME. The data samples used in this study are collected from the Geostationary Operational Environmental Satellite's X-ray flare catalogs provided by the National Centers for Environmental Information. We select M- and X-class flares with identified ARs in the catalogs for the period between 2010 and 2021, and find the associations of flares, CMEs and SEPs in the Space Weather Database of Notifications, Knowledge, Information during the same period. Each data sample contains physical parameters collected from the Helioseismic and Magnetic Imager on board the Solar Dynamics Observatory. Experimental results based on different performance metrics demonstrate that the proposed biLSTM network is better than related machine learning algorithms for the two SEP prediction tasks studied here. We also discuss extensions of our approach for probabilistic forecasting and calibration with empirical evaluation.

Lijie Wang, Yaozong Shen, Shuyuan Peng et al.

While there is increasing concern about the interpretability of neural models, the evaluation of interpretability remains an open problem, due to the lack of proper evaluation datasets and metrics. In this paper, we present a novel benchmark to evaluate the interpretability of both neural models and saliency methods. This benchmark covers three representative NLP tasks: sentiment analysis, textual similarity and reading comprehension, each provided with both English and Chinese annotated data. In order to precisely evaluate the interpretability, we provide token-level rationales that are carefully annotated to be sufficient, compact and comprehensive. We also design a new metric, i.e., the consistency between the rationales before and after perturbations, to uniformly evaluate the interpretability on different types of tasks. Based on this benchmark, we conduct experiments on three typical models with three saliency methods, and unveil their strengths and weakness in terms of interpretability. We will release this benchmark https://www.luge.ai/#/luge/task/taskDetail?taskId=15 and hope it can facilitate the research in building trustworthy systems.

Hao Liu, Matei Zaharia, Pieter Abbeel

Training large transformers using next-token prediction has given rise to groundbreaking advancements in AI. While this generative AI approach has produced impressive results, it heavily leans on human supervision. Even state-of-the-art AI models like ChatGPT depend on fine-tuning through human demonstrations, demanding extensive human input and domain expertise. This strong reliance on human oversight poses a significant hurdle to the advancement of AI innovation. To address this limitation, we propose a novel paradigm termed Exploratory AI (EAI) aimed at autonomously generating high-quality training data. Drawing inspiration from unsupervised reinforcement learning (RL) pretraining, EAI achieves exploration within the natural language space. We accomplish this by harnessing large language models to assess the novelty of generated content. Our approach employs two key components: an actor that generates novel content following exploration principles and a critic that evaluates the generated content, offering critiques to guide the actor. Empirical evaluations demonstrate that EAI significantly boosts model performance on complex reasoning tasks, addressing the limitations of human-intensive supervision.

Arnav Gudibande, Eric Wallace, Charlie Snell et al.

An emerging method to cheaply improve a weaker language model is to finetune it on outputs from a stronger model, such as a proprietary system like ChatGPT (e.g., Alpaca, Self-Instruct, and others). This approach looks to cheaply imitate the proprietary model's capabilities using a weaker open-source model. In this work, we critically analyze this approach. We first finetune a series of LMs that imitate ChatGPT using varying base model sizes (1.5B--13B), data sources, and imitation data amounts (0.3M--150M tokens). We then evaluate the models using crowd raters and canonical NLP benchmarks. Initially, we were surprised by the output quality of our imitation models -- they appear far better at following instructions, and crowd workers rate their outputs as competitive with ChatGPT. However, when conducting more targeted automatic evaluations, we find that imitation models close little to none of the gap from the base LM to ChatGPT on tasks that are not heavily supported in the imitation data. We show that these performance discrepancies may slip past human raters because imitation models are adept at mimicking ChatGPT's style but not its factuality. Overall, we conclude that model imitation is a false promise: there exists a substantial capabilities gap between open and closed LMs that, with current methods, can only be bridged using an unwieldy amount of imitation data or by using more capable base LMs. In turn, we argue that the highest leverage action for improving open-source models is to tackle the difficult challenge of developing better base LMs, rather than taking the shortcut of imitating proprietary systems.

Hao Tian, Can Gao, Xinyan Xiao et al.

Recently, sentiment analysis has seen remarkable advance with the help of pre-training approaches. However, sentiment knowledge, such as sentiment words and aspect-sentiment pairs, is ignored in the process of pre-training, despite the fact that they are widely used in traditional sentiment analysis approaches. In this paper, we introduce Sentiment Knowledge Enhanced Pre-training (SKEP) in order to learn a unified sentiment representation for multiple sentiment analysis tasks. With the help of automatically-mined knowledge, SKEP conducts sentiment masking and constructs three sentiment knowledge prediction objectives, so as to embed sentiment information at the word, polarity and aspect level into pre-trained sentiment representation. In particular, the prediction of aspect-sentiment pairs is converted into multi-label classification, aiming to capture the dependency between words in a pair. Experiments on three kinds of sentiment tasks show that SKEP significantly outperforms strong pre-training baseline, and achieves new state-of-the-art results on most of the test datasets. We release our code at https://github.com/baidu/Senta.

Hongxuan Tang, Hao Liu, Xinyan Xiao

We introduce UGen, a unified autoregressive multimodal model that demonstrates strong performance across text processing, image understanding, and image generation tasks simultaneously. UGen converts both texts and images into discrete token sequences and utilizes a single transformer to generate them uniformly in an autoregressive manner. To address the challenges associated with unified multimodal learning, UGen is trained using a novel mechanism, namely progressive vocabulary learning. In this process, visual token IDs are incrementally activated and integrated into the training phase, ultimately enhancing the effectiveness of unified multimodal learning. Experiments on comprehensive text and image tasks show that UGen achieves a significant overall performance improvement of 13.3% compared to the vanilla unified autoregressive method, and it also delivers competitive results across all tasks against several task-specific models.

Blessing Austin-Gabriel, Aparna S. Varde, Hao Liu

This paper proposes an approach in the area of Knowledge-Guided Machine Learning (KGML) via a novel integrated framework comprising CNN (Convolutional Neural Networks) and ViT (Vision Transformers) along with GIS (Geographic Information Systems) to enhance power plant classification in the context of energy management. Knowledge from geoinformatics derived through Spatial Masks (SM) in GIS is infused into an architecture of CNN and ViT, in this proposed KGML approach. It is found to provide much better performance compared to the baseline of CNN and ViT only in the classification of multiple types of power plants from real satellite imagery, hence emphasizing the vital role of the geoinformatics-guided approach. This work makes a contribution to the main theme of KGML that can be beneficial in many AI systems today. It makes broader impacts on AI in Smart Cities, and Environmental Computing.

Hongxuan Tang, Hao Liu, Xinyan Xiao et al.

Recently, multimodal sentiment analysis has seen remarkable advance and a lot of datasets are proposed for its development. In general, current multimodal sentiment analysis datasets usually follow the traditional system of sentiment/emotion, such as positive, negative and so on. However, when applied in the scenario of video recommendation, the traditional sentiment/emotion system is hard to be leveraged to represent different contents of videos in the perspective of visual senses and language understanding. Based on this, we propose a multimodal sentiment analysis dataset, named baiDu Video Sentiment dataset (DuVideoSenti), and introduce a new sentiment system which is designed to describe the sentimental style of a video on recommendation scenery. Specifically, DuVideoSenti consists of 5,630 videos which displayed on Baidu, each video is manually annotated with a sentimental style label which describes the user's real feeling of a video. Furthermore, we propose UNIMO as our baseline for DuVideoSenti. Experimental results show that DuVideoSenti brings new challenges to multimodal sentiment analysis, and could be used as a new benchmark for evaluating approaches designed for video understanding and multimodal fusion. We also expect our proposed DuVideoSenti could further improve the development of multimodal sentiment analysis and its application to video recommendations.

Hao Liu, Yan Xu, Jiasheng Wang et al.

We propose a new machine learning approach to Stokes inversion based on a convolutional neural network (CNN) and the Milne-Eddington (ME) method. The Stokes measurements used in this study were taken by the Near InfraRed Imaging Spectropolarimeter (NIRIS) on the 1.6 m Goode Solar Telescope (GST) at the Big Bear Solar Observatory. By learning the latent patterns in the training data prepared by the physics-based ME tool, the proposed CNN method is able to infer vector magnetic fields from the Stokes profiles of GST/NIRIS. Experimental results show that our CNN method produces smoother and cleaner magnetic maps than the widely used ME method. Furthermore, the CNN method is 4~6 times faster than the ME method, and is able to produce vector magnetic fields in near real-time, which is essential to space weather forecasting. Specifically, it takes ~50 seconds for the CNN method to process an image of 720 x 720 pixels comprising Stokes profiles of GST/NIRIS. Finally, the CNN-inferred results are highly correlated to the ME-calculated results and are closer to the ME's results with the Pearson product-moment correlation coefficient (PPMCC) being closer to 1 on average than those from other machine learning algorithms such as multiple support vector regression and multilayer perceptrons (MLP). In particular, the CNN method outperforms the current best machine learning method (MLP) by 2.6% on average in PPMCC according to our experimental study. Thus, the proposed physics-assisted deep learning-based CNN tool can be considered as an alternative, efficient method for Stokes inversion for high resolution polarimetric observations obtained by GST/NIRIS.

Hao Liu, Chang Liu, Jason T. L. Wang et al.

We present two recurrent neural networks (RNNs), one based on gated recurrent units and the other based on long short-term memory, for predicting whether an active region (AR) that produces an M- or X-class flare will also produce a coronal mass ejection (CME). We model data samples in an AR as time series and use the RNNs to capture temporal information of the data samples. Each data sample has 18 physical parameters, or features, derived from photospheric vector magnetic field data taken by the Helioseismic and Magnetic Imager (HMI) on board the Solar Dynamics Observatory (SDO). We survey M- and X-class flares that occurred from 2010 May to 2019 May using the Geostationary Operational Environmental Satellite's X-ray flare catalogs provided by the National Centers for Environmental Information (NCEI), and select those flares with identified ARs in the NCEI catalogs. In addition, we extract the associations of flares and CMEs from the Space Weather Database Of Notifications, Knowledge, Information (DONKI). We use the information gathered above to build the labels (positive versus negative) of the data samples at hand. Experimental results demonstrate the superiority of our RNNs over closely related machine learning methods in predicting the labels of the data samples. We also discuss an extension of our approach to predict a probabilistic estimate of how likely an M- or X-class flare will initiate a CME, with good performance results. To our knowledge this is the first time that RNNs have been used for CME prediction.

Hao Liu, Chang Liu, Jason T. L. Wang et al.

We present a long short-term memory (LSTM) network for predicting whether an active region (AR) would produce a gamma-class flare within the next 24 hours. We consider three gamma classes, namely >=M5.0 class, >=M class, and >=C class, and build three LSTM models separately, each corresponding to a gamma class. Each LSTM model is used to make predictions of its corresponding gamma-class flares. The essence of our approach is to model data samples in an AR as time series and use LSTMs to capture temporal information of the data samples. Each data sample has 40 features including 25 magnetic parameters obtained from the Space-weather HMI Active Region Patches (SHARP) and related data products as well as 15 flare history parameters. We survey the flare events that occurred from 2010 May to 2018 May, using the GOES X-ray flare catalogs provided by the National Centers for Environmental Information (NCEI), and select flares with identified ARs in the NCEI flare catalogs. These flare events are used to build the labels (positive vs. negative) of the data samples. Experimental results show that (i) using only 14-22 most important features including both flare history and magnetic parameters can achieve better performance than using all the 40 features together; (ii) our LSTM network outperforms related machine learning methods in predicting the labels of the data samples. To our knowledge, this is the first time that LSTMs have been used for solar flare prediction.