The advancement of socially-aware autonomous vehicles hinges on precise modeling of human behavior. Within this broad paradigm, the specific challenge lies in accurately predicting pedestrian's trajectory and intention. Traditional methodologies have leaned heavily on historical trajectory data, frequently overlooking vital contextual cues such as pedestrian-specific traits and environmental factors. Furthermore, there's a notable knowledge gap as trajectory and intention prediction have largely been approached as separate problems, despite their mutual dependence. To bridge this gap, we introduce PTINet (Pedestrian Trajectory and Intention Prediction Network), which jointly learns the trajectory and intention prediction by combining past trajectory observations, local contextual features (individual pedestrian behaviors), and global features (signs, markings etc.). The efficacy of our approach is evaluated on widely used public datasets: JAAD and PIE, where it has demonstrated superior performance over existing state-of-the-art models in trajectory and intention prediction. The results from our experiments and ablation studies robustly validate PTINet's effectiveness in jointly exploring intention and trajectory prediction for pedestrian behaviour modelling. The experimental evaluation indicates the advantage of using global and local contextual features for pedestrian trajectory and intention prediction. The effectiveness of PTINet in predicting pedestrian behavior paves the way for the development of automated systems capable of seamlessly interacting with pedestrians in urban settings.
In this paper, we describe the results of the HSEmotion team in two tasks of the seventh Affective Behavior Analysis in-the-wild (ABAW) competition, namely, multi-task learning for simultaneous prediction of facial expression, valence, arousal, and detection of action units, and compound expression recognition. We propose an efficient pipeline based on frame-level facial feature extractors pre-trained in multi-task settings to estimate valence-arousal and basic facial expressions given a facial photo. We ensure the privacy-awareness of our techniques by using the lightweight architectures of neural networks, such as MT-EmotiDDAMFN, MT-EmotiEffNet, and MT-EmotiMobileFaceNet, that can run even on a mobile device without the need to send facial video to a remote server. It was demonstrated that a significant step in improving the overall accuracy is the smoothing of neural network output scores using Gaussian or box filters. It was experimentally demonstrated that such a simple post-processing of predictions from simple blending of two top visual models improves the F1-score of facial expression recognition up to 7%. At the same time, the mean Concordance Correlation Coefficient (CCC) of valence and arousal is increased by up to 1.25 times compared to each model's frame-level predictions. As a result, our final performance score on the validation set from the multi-task learning challenge is 4.5 times higher than the baseline (1.494 vs 0.32).
Concurrent processing of multiple autonomous driving 3D perception tasks within the same spatiotemporal scene poses a significant challenge, in particular due to the computational inefficiencies and feature competition between tasks when using traditional multi-task learning approaches. This paper addresses these issues by proposing a novel unified representation, RepVF, which harmonizes the representation of various perception tasks such as 3D object detection and 3D lane detection within a single framework. RepVF characterizes the structure of different targets in the scene through a vector field, enabling a single-head, multi-task learning model that significantly reduces computational redundancy and feature competition. Building upon RepVF, we introduce RFTR, a network designed to exploit the inherent connections between different tasks by utilizing a hierarchical structure of queries that implicitly model the relationships both between and within tasks. This approach eliminates the need for task-specific heads and parameters, fundamentally reducing the conflicts inherent in traditional multi-task learning paradigms. We validate our approach by combining labels from the OpenLane dataset with the Waymo Open dataset. Our work presents a significant advancement in the efficiency and effectiveness of multi-task perception in autonomous driving, offering a new perspective on handling multiple 3D perception tasks synchronously and in parallel. The code will be available at: https://github.com/jbji/RepVF
In recent years, task arithmetic has garnered increasing attention. This approach edits pre-trained models directly in weight space by combining the fine-tuned weights of various tasks into a unified model. Its efficiency and cost-effectiveness stem from its training-free combination, contrasting with traditional methods that require model training on large datasets for multiple tasks. However, applying such a unified model to individual tasks can lead to interference from other tasks (lack of weight disentanglement). To address this issue, Neural Tangent Kernel (NTK) linearization has been employed to leverage a "kernel behavior", facilitating weight disentanglement and mitigating adverse effects from unrelated tasks. Despite its benefits, NTK linearization presents drawbacks, including doubled training costs, as well as reduced performance of individual models. To tackle this problem, we propose a simple yet effective and efficient method that is to finetune the attention modules only in the Transformer. Our study reveals that the attention modules exhibit kernel behavior, and fine-tuning the attention modules only significantly improves weight disentanglement. To further understand how our method improves the weight disentanglement of task arithmetic, we present a comprehensive study of task arithmetic by differentiating the role of the representation module and task-specific module. In particular, we find that the representation module plays an important role in improving weight disentanglement whereas the task-specific modules such as the classification heads can degenerate the weight disentanglement performance. (The code is available at https://github.com/kyrie-23/task_arithmetic_tangent)
Multi-instrument music transcription aims to convert polyphonic music recordings into musical scores assigned to each instrument. This task is challenging for modeling as it requires simultaneously identifying multiple instruments and transcribing their pitch and precise timing, and the lack of fully annotated data adds to the training difficulties. This paper introduces YourMT3+, a suite of models for enhanced multi-instrument music transcription based on the recent language token decoding approach of MT3. We enhance its encoder by adopting a hierarchical attention transformer in the time-frequency domain and integrating a mixture of experts. To address data limitations, we introduce a new multi-channel decoding method for training with incomplete annotations and propose intra- and cross-stem augmentation for dataset mixing. Our experiments demonstrate direct vocal transcription capabilities, eliminating the need for voice separation pre-processors. Benchmarks across ten public datasets show our models' competitiveness with, or superiority to, existing transcription models. Further testing on pop music recordings highlights the limitations of current models. Fully reproducible code and datasets are available with demos at \url{https://github.com/mimbres/YourMT3}.
Split Computing (SC), where a Deep Neural Network (DNN) is intelligently split with a part of it deployed on an edge device and the rest on a remote server is emerging as a promising approach. It allows the power of DNNs to be leveraged for latency-sensitive applications that do not allow the entire DNN to be deployed remotely, while not having sufficient computation bandwidth available locally. In many such embedded systems scenarios, such as those in the automotive domain, computational resource constraints also necessitate Multi-Task Learning (MTL), where the same DNN is used for multiple inference tasks instead of having dedicated DNNs for each task, which would need more computing bandwidth. However, how to partition such a multi-tasking DNN to be deployed within a SC framework has not been sufficiently studied. This paper studies this problem, and MTL-Split, our novel proposed architecture, shows encouraging results on both synthetic and real-world data. The source code is available at https://github.com/intelligolabs/MTL-Split.
Scoliosis presents significant diagnostic challenges, particularly in adolescents, where early detection is crucial for effective treatment. Traditional diagnostic and follow-up methods, which rely on physical examinations and radiography, face limitations due to the need for clinical expertise and the risk of radiation exposure, thus restricting their use for widespread early screening. In response, we introduce a novel video-based, non-invasive method for scoliosis classification using gait analysis, effectively circumventing these limitations. This study presents Scoliosis1K, the first large-scale dataset specifically designed for video-based scoliosis classification, encompassing over one thousand adolescents. Leveraging this dataset, we developed ScoNet, an initial model that faced challenges in handling the complexities of real-world data. This led to the development of ScoNet-MT, an enhanced model incorporating multi-task learning, which demonstrates promising diagnostic accuracy for practical applications. Our findings demonstrate that gait can serve as a non-invasive biomarker for scoliosis, revolutionizing screening practices through deep learning and setting a precedent for non-invasive diagnostic methodologies. The dataset and code are publicly available at https://zhouzi180.github.io/Scoliosis1K/.
We aim to address Multi-Task Learning (MTL) with a large number of tasks by Multi-Task Grouping (MTG). Given N tasks, we propose to simultaneously identify the best task groups from 2^N candidates and train the model weights simultaneously in one-shot, with the high-order task-affinity fully exploited. This is distinct from the pioneering methods which sequentially identify the groups and train the model weights, where the group identification often relies on heuristics. As a result, our method not only improves the training efficiency, but also mitigates the objective bias introduced by the sequential procedures that potentially lead to a suboptimal solution. Specifically, we formulate MTG as a fully differentiable pruning problem on an adaptive network architecture determined by an underlying Categorical distribution. To categorize N tasks into K groups (represented by K encoder branches), we initially set up KN task heads, where each branch connects to all N task heads to exploit the high-order task-affinity. Then, we gradually prune the KN heads down to N by learning a relaxed differentiable Categorical distribution, ensuring that each task is exclusively and uniquely categorized into only one branch. Extensive experiments on CelebA and Taskonomy datasets with detailed ablations show the promising performance and efficiency of our method. The codes are available at https://github.com/ethanygao/DMTG.
This paper introduces a novel combination of two tasks, previously treated separately: acoustic-to-articulatory speech inversion (AAI) and phoneme-to-articulatory (PTA) motion estimation. We refer to this joint task as acoustic phoneme-to-articulatory speech inversion (APTAI) and explore two different approaches, both working speaker- and text-independently during inference. We use a multi-task learning setup, with the end-to-end goal of taking raw speech as input and estimating the corresponding articulatory movements, phoneme sequence, and phoneme alignment. While both proposed approaches share these same requirements, they differ in their way of achieving phoneme-related predictions: one is based on frame classification, the other on a two-staged training procedure and forced alignment. We reach competitive performance of 0.73 mean correlation for the AAI task and achieve up to approximately 87% frame overlap compared to a state-of-the-art text-dependent phoneme force aligner.
Efficient machine learning (ML) has become increasingly important as models grow larger and data volumes expand. In this work, we address the trade-off between generalization in multi-task learning (MTL) and precision in single-task learning (STL) by introducing the Multi-Task to Single-Task (MT2ST) framework. MT2ST is designed to enhance training efficiency and accuracy in word embedding tasks, showcasing its value as a practical application of efficient ML. Our framework employs two strategies: *Diminish*, which gradually reduces the influence of auxiliary tasks, and *Switch*, which transitions training from MTL to STL at a specific point. Empirical results show that MT2ST reduces training time by 67\% compared to STL and by 13\% compared to traditional MTL, while maintaining high accuracy. These findings highlight MT2ST as an efficient ML solution tailored for optimizing word embedding training. Code is available at https://github.com/NoakLiu/MT2ST.
Urbanization advances at unprecedented rates, resulting in negative effects on the environment and human well-being. Remote sensing has the potential to mitigate these effects by supporting sustainable development strategies with accurate information on urban growth. Deep learning-based methods have achieved promising urban change detection results from optical satellite image pairs using convolutional neural networks (ConvNets), transformers, and a multi-task learning setup. However, transformers have not been leveraged for urban change detection with multi-temporal data, i.e., >2 images, and multi-task learning methods lack integration approaches that combine change and segmentation outputs. To fill this research gap, we propose a continuous urban change detection method that identifies changes in each consecutive image pair of a satellite image time series (SITS). Specifically, we propose a temporal feature refinement (TFR) module that utilizes self-attention to improve ConvNet-based multi-temporal building representations. Furthermore, we propose a multi-task integration (MTI) module that utilizes Markov networks to find an optimal building map time series based on segmentation and dense change outputs. The proposed method effectively identifies urban changes based on high-resolution SITS acquired by the PlanetScope constellation (F1 score 0.551) and Gaofen-2 (F1 score 0.440). Moreover, our experiments on two challenging datasets demonstrate the effectiveness of the proposed method compared to bi-temporal and multi-temporal urban change detection and segmentation methods.
Many optimization problems require balancing multiple conflicting objectives. As gradient descent is limited to single-objective optimization, we introduce its direct generalization: Jacobian descent (JD). This algorithm iteratively updates parameters using the Jacobian matrix of a vector-valued objective function, in which each row is the gradient of an individual objective. While several methods to combine gradients already exist in the literature, they are generally hindered when the objectives conflict. In contrast, we propose projecting gradients to fully resolve conflict while ensuring that they preserve an influence proportional to their norm. We prove significantly stronger convergence guarantees with this approach, supported by our empirical results. Our method also enables instance-wise risk minimization (IWRM), a novel learning paradigm in which the loss of each training example is considered a separate objective. Applied to simple image classification tasks, IWRM exhibits promising results compared to the direct minimization of the average loss. Additionally, we outline an efficient implementation of JD using the Gramian of the Jacobian matrix to reduce time and memory requirements.
This paper introduces RouteFinder, a comprehensive foundation model framework to tackle different Vehicle Routing Problem (VRP) variants. Our core idea is that a foundation model for VRPs should be able to represent variants by treating each as a subset of a generalized problem equipped with different attributes. We propose a unified VRP environment capable of efficiently handling any attribute combination. The RouteFinder model leverages a modern transformer-based encoder and global attribute embeddings to improve task representation. Additionally, we introduce two reinforcement learning techniques to enhance multi-task performance: mixed batch training, which enables training on different variants at once, and multi-variant reward normalization to balance different reward scales. Finally, we propose efficient adapter layers that enable fine-tuning for new variants with unseen attributes. Extensive experiments on 24 VRP variants show RouteFinder achieves competitive results. Our code is openly available at https://github.com/ai4co/routefinder.
Training a neural network-based biomedical named entity recognition (BioNER) model usually requires extensive and costly human annotations. While several studies have employed multi-task learning with multiple BioNER datasets to reduce human effort, this approach does not consistently yield performance improvements and may introduce label ambiguity in different biomedical corpora. We aim to tackle those challenges through transfer learning from easily accessible resources with fewer concept overlaps with biomedical datasets. We proposed GERBERA, a simple-yet-effective method that utilized general-domain NER datasets for training. We performed multi-task learning to train a pre-trained biomedical language model with both the target BioNER dataset and the general-domain dataset. Subsequently, we fine-tuned the models specifically for the BioNER dataset. We systematically evaluated GERBERA on five datasets of eight entity types, collectively consisting of 81,410 instances. Despite using fewer biomedical resources, our models demonstrated superior performance compared to baseline models trained with additional BioNER datasets. Specifically, our models consistently outperformed the baseline models in six out of eight entity types, achieving an average improvement of 0.9% over the best baseline performance across eight entities. Our method was especially effective in amplifying performance on BioNER datasets characterized by limited data, with a 4.7% improvement in F1 scores on the JNLPBA-RNA dataset. This study introduces a new training method that leverages cost-effective general-domain NER datasets to augment BioNER models. This approach significantly improves BioNER model performance, making it a valuable asset for scenarios with scarce or costly biomedical datasets.
Learning predictive world models is crucial for enhancing the planning capabilities of reinforcement learning (RL) agents. Recently, MuZero-style algorithms, leveraging the value equivalence principle and Monte Carlo Tree Search (MCTS), have achieved superhuman performance in various domains. However, these methods struggle to scale in heterogeneous scenarios with diverse dependencies and task variability. To overcome these limitations, we introduce UniZero, a novel approach that employs a modular transformer-based world model to effectively learn a shared latent space. By concurrently predicting latent dynamics and decision-oriented quantities conditioned on the learned latent history, UniZero enables joint optimization of the long-horizon world model and policy, facilitating broader and more efficient planning in the latent space. We show that UniZero significantly outperforms existing baselines in benchmarks that require long-term memory. Additionally, UniZero demonstrates superior scalability in multitask learning experiments conducted on Atari benchmarks. In standard single-task RL settings, such as Atari and DMControl, UniZero matches or even surpasses the performance of current state-of-the-art methods. Finally, extensive ablation studies and visual analyses validate the effectiveness and scalability of UniZero's design choices. Our code is available at \textcolor{magenta}{https://github.com/opendilab/LightZero}.
Semantic change detection is an important task in geoscience and earth observation. By producing a semantic change map for each temporal phase, both the land use land cover categories and change information can be interpreted. Recently some multi-task learning based semantic change detection methods have been proposed to decompose the task into semantic segmentation and binary change detection subtasks. However, previous works comprise triple branches in an entangled manner, which may not be optimal and hard to adopt foundation models. Besides, lacking explicit refinement of bitemporal features during fusion may cause low accuracy. In this letter, we propose a novel late-stage bitemporal feature fusion network to address the issue. Specifically, we propose local global attentional aggregation module to strengthen feature fusion, and propose local global context enhancement module to highlight pivotal semantics. Comprehensive experiments are conducted on two public datasets, including SECOND and Landsat-SCD. Quantitative and qualitative results show that our proposed model achieves new state-of-the-art performance on both datasets.
Solving multi-objective optimization problems for large deep neural networks is a challenging task due to the complexity of the loss landscape and the expensive computational cost of training and evaluating models. Efficient Pareto front approximation of large models enables multi-objective optimization for various tasks such as multi-task learning and trade-off analysis. Existing algorithms for learning Pareto set, including (1) evolutionary, hypernetworks, and hypervolume-maximization methods, are computationally expensive and have restricted scalability to large models; (2) Scalarization algorithms, where a separate model is trained for each objective ray, which is inefficient for learning the entire Pareto set and fails to capture the objective trade-offs effectively. Inspired by the recent success of model merging, we propose a practical and scalable approach to Pareto set learning problem via mixture of experts (MoE) based model fusion. By ensembling the weights of specialized single-task models, the MoE module can effectively capture the trade-offs between multiple objectives and closely approximate the entire Pareto set of large neural networks. Once the routers are learned and a preference vector is set, the MoE module can be unloaded, thus no additional computational cost is introduced during inference. We conduct extensive experiments on vision and language tasks using large-scale models such as CLIP-ViT and GPT-2. The experimental results demonstrate that our method efficiently approximates the entire Pareto front of large models. Using only hundreds of trainable parameters of the MoE routers, our method even has lower memory usage compared to linear scalarization and algorithms that learn a single Pareto optimal solution, and are scalable to both the number of objectives and the size of the model.
This paper introduces DeepUnifiedMom, a deep learning framework that enhances portfolio management through a multi-task learning approach and a multi-gate mixture of experts. The essence of DeepUnifiedMom lies in its ability to create unified momentum portfolios that incorporate the dynamics of time series momentum across a spectrum of time frames, a feature often missing in traditional momentum strategies. Our comprehensive backtesting, encompassing diverse asset classes such as equity indexes, fixed income, foreign exchange, and commodities, demonstrates that DeepUnifiedMom consistently outperforms benchmark models, even after factoring in transaction costs. This superior performance underscores DeepUnifiedMom's capability to capture the full spectrum of momentum opportunities within financial markets. The findings highlight DeepUnifiedMom as an effective tool for practitioners looking to exploit the entire range of momentum opportunities. It offers a compelling solution for improving risk-adjusted returns and is a valuable strategy for navigating the complexities of portfolio management.
Multi-task learning (MTL) is a powerful machine learning paradigm designed to leverage shared knowledge across tasks to improve generalization and performance. Previous works have proposed approaches to MTL that can be divided into feature learning, focused on the identification of a common feature representation, and task clustering, where similar tasks are grouped together. In this paper, we propose an MTL approach at the intersection between task clustering and feature transformation based on a two-phase iterative aggregation of targets and features. First, we propose a bias-variance analysis for regression models with additive Gaussian noise, where we provide a general expression of the asymptotic bias and variance of a task, considering a linear regression trained on aggregated input features and an aggregated target. Then, we exploit this analysis to provide a two-phase MTL algorithm (NonLinCTFA). Firstly, this method partitions the tasks into clusters and aggregates each obtained group of targets with their mean. Then, for each aggregated task, it aggregates subsets of features with their mean in a dimensionality reduction fashion. In both phases, a key aspect is to preserve the interpretability of the reduced targets and features through the aggregation with the mean, which is further motivated by applications to Earth science. Finally, we validate the algorithms on synthetic data, showing the effect of different parameters and real-world datasets, exploring the validity of the proposed methodology on classical datasets, recent baselines, and Earth science applications.
Learning general-purpose models from diverse datasets has achieved great success in machine learning. In robotics, however, existing methods in multi-task learning are typically constrained to a single robot and workspace, while recent work such as RT-X requires a non-trivial action normalization procedure to manually bridge the gap between different action spaces in diverse environments. In this paper, we propose the visual kinematics chain as a precise and universal representation of quasi-static actions for robot learning over diverse environments, which requires no manual adjustment since the visual kinematic chains can be automatically obtained from the robot's model and camera parameters. We propose the Visual Kinematics Transformer (VKT), a convolution-free architecture that supports an arbitrary number of camera viewpoints, and that is trained with a single objective of forecasting kinematic structures through optimal point-set matching. We demonstrate the superior performance of VKT over BC transformers as a general agent on Calvin, RLBench, Open-X, and real robot manipulation tasks. Video demonstrations can be found at https://mlzxy.github.io/visual-kinetic-chain.