Zero-shot Transfer Research Articles

Multilingual Auxiliary Tasks Training: Bridging the Gap between Languages for Zero-Shot Transfer of Hate Speech Detection Models

Multilingual Auxiliary Tasks Training: Bridging the Gap between Languages for Zero-Shot Transfer of Hate Speech Detection Models

TC-Driver: A Trajectory Conditioned Reinforcement Learning Approach to Zero-Shot Autonomous Racing

Autonomous racing is becoming popular for academic and industry researchers as a test for general autonomous driving by pushing perception, planning, and control algorithms to their limits. While traditional control methods such as model predictive control are capable of generating an optimal control sequence at the edge of the vehicles’ physical controllability, these methods are sensitive to the accuracy of the modeling parameters, such as tire modeling coefficients. As model mismatch is inevitable in reality, the heuristic nature of Reinforcement Learning (RL) offers a viable approach to modeling robustness. This paper presents TC-Driver, an RL approach for robust control in autonomous racing. In particular, the TC-Driver agent is conditioned by a trajectory generated by any arbitrary traditional high-level trajectory planner. The proposed TC-Driver architecture addresses the tire parameter modeling inaccuracies by exploiting the learning capabilities of RL while utilizing the reliability of traditional planning methods in a hybrid fashion. We train the agent under varying tire conditions, allowing it to generalize to different model parameters, aiming to increase the racing capabilities of the system in practice. Experimental results demonstrate that the proposed hybrid RL architecture of the TC-Driver improves the generalization robustness of autonomous racing agents when compared to a previous state-of-the-art end-to-end-based architecture. Namely, the proposed controller yields a 29-fold improvement in crash ratio when facing model mismatch and can zero-shot transfer its behavior on unseen tracks which present completely new features, while the end-to-end baseline fails. When deployed on a physical system, the proposed architecture demonstrates zero-shot Sim2Real capabilities that outperform end-to-end agents 10-fold in terms of crash ratio while exhibiting similar driving characteristics in reality as in simulation.

MSeg: A Composite Dataset for Multi-Domain Semantic Segmentation.

We present MSeg, a composite dataset that unifies semantic segmentation datasets from different domains. A naive merge of the constituent datasets yields poor performance due to inconsistent taxonomies and annotation practices. We reconcile the taxonomies and bring the pixel-level annotations into alignment by relabeling more than 220,000 object masks in more than 80,000 images, requiring more than 1.34 years of collective annotator effort. The resulting composite dataset enables training a single semantic segmentation model that functions effectively across domains and generalizes to datasets that were not seen during training. We adopt zero-shot cross-dataset transfer as a benchmark to systematically evaluate a model's robustness and show that MSeg training yields substantially more robust models in comparison to training on individual datasets or naive mixing of datasets without the presented contributions. A model trained on MSeg ranks first on the WildDash-v1 leaderboard for robust semantic segmentation, with no exposure to WildDash data during training. We evaluate our models in the 2020 Robust Vision Challenge (RVC) as an extreme generalization experiment. MSeg training sets include only three of the seven datasets in the RVC; more importantly, the evaluation taxonomy of RVC is different and more detailed. Surprisingly, our model shows competitive performance and ranks second. To evaluate how close we are to the grand aim of robust, efficient, and complete scene understanding, we go beyond semantic segmentation by training instance segmentation and panoptic segmentation models using our dataset. Moreover, we also evaluate various engineering design decisions and metrics, including resolution and computational efficiency. Although our models are far from this grand aim, our comprehensive evaluation is crucial for progress. We share all the models and code with the community.

The Sentiment of Latin Poetry. Annotation and Automatic Analysis of the Odes of Horace

During the recent years, an always growing number of linguistic resources and automatic systems for sentiment analysis have been developed covering a wide range of languages. However, research in this field is still not much explored for texts written in Classical languages. Working on such languages means dealing with peculiar textual genres such as philosophical, historical or religious treatises, epic narratives, plays and poems. Poems are particularly suitable for sentiment analysis because they tell us about emotions and passions. In this paper, we describe the creation of the first small gold standard of Latin made of poems written by Horace and manually annotated with emotion polarity, but we also report about the results of a set of automatic classification experiments. In particular, we test both a lexicon-based approach, which uses a Latin polarity lexicon called LatinAffectus, and a zero-shot transfer method. We provide details about the methodology adopted for the annotation of the gold standard, the creation of LatinAffectus, the development of our experiments and we give details about the results and the limitations of the proposed approaches.

Language-Aware Domain Generalization Network for Cross-Scene Hyperspectral Image Classification

Text information including extensive prior knowledge about land cover classes has been ignored in hyperspectral image classification (HSI) tasks. It is necessary to explore the effectiveness of linguistic mode in assisting HSI classification. In addition, the large-scale pre-training image-text foundation models have demonstrated great performance in a variety of downstream applications, including zero-shot transfer. However, most domain generalization methods have never addressed mining linguistic modal knowledge to improve the generalization performance of model. To compensate for the inadequacies listed above, a Language-aware Domain Generalization Network (LDGnet) is proposed to learn cross-domain invariant representation from cross-domain shared prior knowledge. The proposed method only trains on the source domain (SD) and then transfers the model to the target domain (TD). The dual-stream architecture including image encoder and text encoder is used to extract visual and linguistic features, in which coarse-grained and fine-grained text representations are designed to extract two levels of linguistic features. Furthermore, linguistic features are used as cross-domain shared semantic space, and visual-linguistic alignment is completed by supervised contrastive learning in semantic space. Extensive experiments on three datasets demonstrate the superiority of the proposed method when compared with state-of-the-art techniques.

Answer Prediction for Questions from Tamil and Hindi Passages

Predicting answers for questions based on passages from Tamil and Hindi is a Question Answering (QA) task. It helps us to elicit information that is relevant to the queries from the passage. Answers are predicted by identifying the answer span from the context based on the questions. In this paper, four different approaches are used to solve the QA task for Tamil and Hindi Passages. Zero-shot transfer of QA models has been tested on the QA datasets available for Hindi and Tamil. Comparative analysis between Multilingual models that are fine-tuned using Tamil and Hindi QA datasets is carried out. In addition to that, the performance has also been evaluated after increasing the Hindi data points. As there are very few QA datasets available in Indic Languages, especially Tamil, a possible solution to overcome this challenge was tried and tested using the creation of a QA dataset for Tamil using Indic Language Translation models. Multilingual models were then used for fine-tuning with the translated Tamil QA dataset generated using Language translation and the results have been discussed. The predicted answers will be evaluated using Jaccard Similarity, Exact match (EM), and F1 Scores.

Zero-Shot Transfer Learning Framework for Plant Leaf Disease Classification

Zero-Shot Transfer Learning Framework for Plant Leaf Disease Classification

FaithDial: A Faithful Benchmark for Information-Seeking Dialogue

Abstract The goal of information-seeking dialogue is to respond to seeker queries with natural language utterances that are grounded on knowledge sources. However, dialogue systems often produce unsupported utterances, a phenomenon known as hallucination. To mitigate this behavior, we adopt a data-centric solution and create FaithDial, a new benchmark for hallucination-free dialogues, by editing hallucinated responses in the Wizard of Wikipedia (WoW) benchmark. We observe that FaithDial is more faithful than WoW while also maintaining engaging conversations. We show that FaithDial can serve as training signal for: i) a hallucination critic, which discriminates whether an utterance is faithful or not, and boosts the performance by 12.8 F1 score on the BEGIN benchmark compared to existing datasets for dialogue coherence; ii) high-quality dialogue generation. We benchmark a series of state-of-the-art models and propose an auxiliary contrastive objective that achieves the highest level of faithfulness and abstractiveness based on several automated metrics. Further, we find that the benefits of FaithDial generalize to zero-shot transfer on other datasets, such as CMU-Dog and TopicalChat. Finally, human evaluation reveals that responses generated by models trained on FaithDial are perceived as more interpretable, cooperative, and engaging.

Using Fuzzy Logic to Learn Abstract Policies in Large-Scale Multiagent Reinforcement Learning

Large-scale multiagent reinforcement learning requires huge computation and space costs, and the too-long execution process makes it hard to train policies for agents. This work proposes a concept of fuzzy agent, which is a new paradigm for training homogeneous agents. Aiming at a lightweight and affordable reinforcement learning mechanism for large-scale homogeneous multiagent systems, we break the one-to-one correspondence between agent and policy, designing abstract agents as the substitute for the multiagent to interact with the environment. The Markov decision process models for these abstract agents are conducted by fuzzy logic, which also acts on the behavior mapping from abstract agent to entity. Specifically, just the abstract agents execute their policy at a time step, and the concrete behaviors are generated by simple matrix operations. The proposal has lower space and computation complexities because the number of abstract agents is far less than that of entities, and the coupling among agents is retained implicitly. Compared with other approximation and simplification methods, the proposed fuzzy agent not only greatly reduces required computing resources but also ensures the effectiveness of the learned policies. Several experiments are conducted to validate our method. The results show that the proposal outperforms the baseline methods, while it has satisfactory zero-shot and few-shot transfer abilities.

Dealing with textual noise for robust and effective BERT re-ranking

The pre-trained language models (PLMs), such as BERT, have been successfully employed in two-phases ranking pipeline for information retrieval (IR). Meanwhile, recent studies have reported that BERT model is vulnerable to imperceptible textual perturbations on quite a few natural language processing (NLP) tasks. As for IR tasks, current established BERT re-ranker is mainly trained on large-scale and relatively clean dataset, such as MS MARCO, but actually noisy text is more common in real-world scenarios, such as web search. In addition, the impact of within-document textual noises (perturbations) on retrieval effectiveness remains to be investigated, especially on the ranking quality of BERT re-ranker, considering its contextualized nature. To mitigate this gap, we carry out exploratory experiments on the MS MARCO dataset in this work to examine whether BERT re-ranker can still perform well when ranking text with noise. Unfortunately, we observe non-negligible effectiveness degradation of BERT re-ranker over a total of ten different types of synthetic within-document textual noise. Furthermore, to address the effectiveness losses over textual noise, we propose a novel noise-tolerant model, De-Ranker, which is learned by minimizing the distance between the noisy text and its original clean version. Our evaluation on the MS MARCO and TREC 2019–2020 DL datasets demonstrates that De-Ranker can deal with synthetic textual noise more effectively, with 3%–4% performance improvement over vanilla BERT re-ranker. Meanwhile, extensive zero-shot transfer experiments on a total of 18 widely-used IR datasets show that De-Ranker can not only tackle natural noise in real-world text, but also achieve 1.32% improvement on average in terms of cross-domain generalization ability on the BEIR benchmark.

Deep-AI soft sensor for sustainable health risk monitoring and control of fine particulate matter at sensor devoid underground spaces: A zero-shot transfer learning approach

Underground building spaces such as metro stations have been widely adopted in densely populated metropolitan cities to combat high traffic congestion. To ensure a sustainable underground environment, early warning systems, primarily for fine particulate matter, and ventilation control, are of great significance. However, these techniques require substantial sensor data, which is generally not accessible in real-life situations. This work presents a solution to sensor data unavailability at underground metro stations by integrating a sequence forecasting soft-sensor framework with zero-shot transfer learning (ZSTL). The proposed data-driven soft-sensor model consists of a multi-head attention aware bi-directional gated recurrent unit (MH-BiGRU) that forecasts the PM2.5 one hour ahead of time to provide an early health risk warning and proactive ventilation operation. The results suggest that the forecasts by the proposed model outperform stacked long short-term memory and gated recurrent unit structures by 17% and 20%, respectively. Whereas 72% of the residual errors by MH-BiGRU were situated between −5 to 5 CIAI units. Additionally, the proposed ZSTL framework reduced the zero-shot prediction error by 75%, 17%, and 82% at three target metro stations, where the data of target sensor PM2.5 was not available. The ventilation control by forecasted PM2.5 levels can decrease indoor PM2.5 concentrations by 25% and has the potential to reduce energy demand by 18%, enabling sustainable operation.

Deep Learning-Based NMPC for Local Motion Planning of Last-Mile Delivery Robot.

Feasible local motion planning for autonomous mobile robots in dynamic environments requires predicting how the scene evolves. Conventional navigation stakes rely on a local map to represent how a dynamic scene changes over time. However, these navigation stakes depend highly on the accuracy of the environmental map and the number of obstacles. This study uses semantic segmentation-based drivable area estimation as an alternative representation to assist with local motion planning. Notably, a realistic 3D simulator based on an Unreal Engine was created to generate a synthetic dataset under different weather conditions. A transfer learning technique was used to train the encoder-decoder model to segment free space from the occupied sidewalk environment. The local planner uses a nonlinear model predictive control (NMPC) scheme that inputs the estimated drivable space, the state of the robot, and a global plan to produce safe velocity commands that minimize the tracking cost and actuator effort while avoiding collisions with dynamic and static obstacles. The proposed approach achieves zero-shot transfer from a simulation to real-world environments that have never been experienced during training. Several intensive experiments were conducted and compared with the dynamic window approach (DWA) to demonstrate the effectiveness of our system in dynamic sidewalk environments.

Reinforcement Learning of Impedance Policies for Peg-in-Hole Tasks: Role of Asymmetric Matrices

Robotic manipulators are playing an increasing role in a wide range of industries. However, their application to assembly tasks is hampered by the need for precise control over the environment and for task-specific coding. Cartesian impedance control is a well-established method for interacting with the environment and handling uncertainties. With the advance of Reinforcement Learning (RL) it has been suggested to learn the impedance matrices. However, most of the current work is limited to learning diagonal impedance matrices in addition to the trajectory itself. We argue that asymmetric impedance matrices enhance the ability to properly correct reference trajectories generated by a baseline planner, alleviating the need for learning the trajectory. Moreover, a task-specific set of asymmetric impedance matrices can be sufficient for simple tasks, alleviating the need for learning variable impedance control. We learn impedance policies for small (few mm) peg-in-hole using model-free RL, and investigate the advantage of using asymmetric impedance matrices and their space-invariance. Finally, we demonstrate zero-shot policy transfer from the simulation to a real robot, and generalization to new real-world environments, with larger parts and semi-flexible pegs.

Reinforcement Learning for Autonomous Mobility-on-Demand Systems

Autonomous mobility-on-demand (AMoD) systems represent a rapidly developing mode of transportation wherein travel requests are dynamically handled by a coordinated fleet of robotic, self-driving vehicles. Given a graph representation of the transportation network - one where, for example, nodes represent areas of the city, and edges the connectivity between them - we argue that the AMoD control problem is naturally cast as a node-wise decision-making problem. In this paper, we propose a deep reinforcement learning framework1 to control the rebalancing of AMoD systems through graph neural networks. Crucially, we demonstrate that graph neural networks enable reinforcement learning agents to recover behavior policies that are significantly more transferable, generalizable, and scalable than policies learned through other approaches. Empirically, we show how the learned policies exhibit promising zero-shot transfer capabilities when faced with critical portability tasks such as inter-city generalization, service area expansion, and adaptation to potentially complex urban topologies.

Learning to Prompt for Vision-Language Models

Large pre-trained vision-language models like CLIP have shown great potential in learning representations that are transferable across a wide range of downstream tasks. Different from the traditional representation learning that is based mostly on discretized labels, vision-language pre-training aligns images and texts in a common feature space, which allows zero-shot transfer to a downstream task via prompting, i.e., classification weights are synthesized from natural language describing classes of interest. In this work, we show that a major challenge for deploying such models in practice is prompt engineering, which requires domain expertise and is extremely time-consuming -- one needs to spend a significant amount of time on words tuning since a slight change in wording could have a huge impact on performance. Inspired by recent advances in prompt learning research in natural language processing (NLP), we propose Context Optimization (CoOp), a simple approach specifically for adapting CLIP-like vision-language models for downstream image recognition. Concretely, CoOp models a prompt's context words with learnable vectors while the entire pre-trained parameters are kept fixed. To handle different image recognition tasks, we provide two implementations of CoOp: unified context and class-specific context. Through extensive experiments on 11 datasets, we demonstrate that CoOp requires as few as one or two shots to beat hand-crafted prompts with a decent margin and is able to gain significant improvements over prompt engineering with more shots, e.g., with 16 shots the average gain is around 15% (with the highest reaching over 45%). Despite being a learning-based approach, CoOp achieves superb domain generalization performance compared with the zero-shot model using hand-crafted prompts.

MRI-based model for MCI conversion using deep zero-shot transfer learning

MRI-based model for MCI conversion using deep zero-shot transfer learning

Autonomous State-Based Flipper Control for Articulated Tracked Robots in Urban Environments

We demonstrate a hybrid approach to autonomous flipper control, focusing on a fusion of hard-coded and learned knowledge. The result is a sample-efficient and modifiable control structure that can be used in conjunction with a mapping/navigation stack. The backbone of the control policy is formulated as a state machine whose states define various flipper action templates and local control behaviors. It is also used as an interface that facilitates the gathering of demonstrations to train the transitions of the state machine. We propose a soft-differentiable state machine neural network that mitigates the shortcomings of its naively implemented counterpart and improves over a multi-layer perceptron baseline in the task of state-transition classification. We show that by training on several minutes of user-gathered demonstrations in simulation, our approach is capable of a zero-shot domain transfer to a wide range of obstacles on a similar real robotic platform. Our results show a considerable increase in performance over a previous competing approach in several essential criteria. A subset of this work was successfully used in the Defense Advanced Research Projects Agency (DARPA) Subterranean Challenge to alleviate the operator of manual flipper control. We autonomously traversed stairs and other obstacles, improving map coverage.

CL-NERIL: A Cross-Lingual Model for NER in Indian Languages (Student Abstract)

Developing Named Entity Recognition (NER) systems for Indian languages has been a long-standing challenge, mainly owing to the requirement of a large amount of annotated clean training instances. This paper proposes an end-to-end framework for NER for Indian languages in a low-resource setting by exploiting parallel corpora of English and Indian languages and an English NER dataset. The proposed framework includes an annotation projection method that combines word alignment score and NER tag prediction confidence score on source language (English) data to generate weakly labeled data in a target Indian language. We employ a variant of the Teacher-Student model and optimize it jointly on the pseudo labels of the Teacher model and predictions on the generated weakly labeled data. We also present manually annotated test sets for three Indian languages: Hindi, Bengali, and Gujarati. We evaluate the performance of the proposed framework on the test sets of the three Indian languages. Empirical results show a minimum 10% performance improvement compared to the zero-shot transfer learning model on all languages. This indicates that weakly labeled data generated using the proposed annotation projection method in target Indian languages can complement well-annotated source language data to enhance performance. Our code is publicly available at https://github.com/aksh555/CL-NERIL.

LITMUS Predictor: An AI Assistant for Building Reliable, High-Performing and Fair Multilingual NLP Systems

Pre-trained multilingual language models are gaining popularity due to their cross-lingual zero-shot transfer ability, but these models do not perform equally well in all languages. Evaluating task-specific performance of a model in a large number of languages is often a challenge due to lack of labeled data, as is targeting improvements in low performing languages through few-shot learning. We present a tool - LITMUS Predictor - that can make reliable performance projections for a fine-tuned task-specific model in a set of languages without test and training data, and help strategize data labeling efforts to optimize performance and fairness objectives.

AdaSL: An Unsupervised Domain Adaptation framework for Arabic multi-dialectal Sequence Labeling

Dialectal Arabic (DA) refers to varieties of everyday spoken languages in the Arab world. These dialects differ according to the country and region of the speaker, and their textual content is constantly growing with the rise of social media networks and web blogs. Although research on Natural Language Processing (NLP) on standard Arabic, namely Modern Standard Arabic (MSA), has witnessed remarkable progress, research efforts on DA are rather limited. This is due to numerous challenges, such as the scarcity of labeled data as well as the nature and structure of DA. While some recent works have reached decent results on several DA sentence classification tasks, other complex tasks, such as sequence labeling, still suffer from weak performances when it comes to DA varieties with either a limited amount of labeled data or unlabeled data only. Besides, it has been shown that zero-shot transfer learning from models trained on MSA does not perform well on DA. In this paper, we introduce AdaSL, a new unsupervised domain adaptation framework for Arabic multi-dialectal sequence labeling, leveraging unlabeled DA data, labeled MSA data, and existing multilingual and Arabic Pre-trained Language Models (PLMs). The proposed framework relies on four key components: (1) domain adaptive fine-tuning of multilingual/MSA language models on unlabeled DA data, (2) sub-word embedding pooling, (3) iterative self-training on unlabeled DA data, and (4) iterative DA and MSA distribution alignment. We evaluate our framework on multi-dialectal Named Entity Recognition (NER) and Part-of-Speech (POS) tagging tasks.The overall results show that the zero-shot transfer learning, using our proposed framework, boosts the performance of the multilingual PLMs by 40.87% in macro-F1 score for the NER task, while it boosts the accuracy by 6.95% for the POS tagging task. For the Arabic PLMs, our proposed framework increases performance by 16.18% macro-F1 for the NER task and 2.22% accuracy for the POS tagging task, and thus, achieving new state-of-the-art zero-shot transfer learning performance for Arabic multi-dialectal sequence labeling.