Further Evidence for the Binding and Retrieval of Control-States From the Flanker Task

In response-interference tasks, congruency effects are reduced in trials that follow an incongruent trial. This congruence sequence effect (CSE) has been taken to reflect top-down cognitive control processes that monitor for and intervene in case of conflict. In contrast, episodic-memory accounts explain CSEs with bottom-up retrieval of stimulus-response links. Reconciling these opposing views, an emerging perspective holds that memory stores instances of control – abstract control-states – creating a shortcut for effortful control processes. Support comes from a study that assessed CSEs in a prime-target task. Here, repeating an irrelevant context feature boosted CSEs, possibly by retrieving previously stored control-states. We present a conceptual replication using the Eriksen flanker task because previous research found that CSEs in the flanker task reflect different control mechanisms than CSEs in the prime-target task. We measured CSEs while controlling for stimulus–response memory effects and manipulated contextual information (vertical spatial location) independently from the stimulus information, which introduced the conflict (horizontal spatial location). Results replicate previous findings – CSEs increased for context-repetition compared to context-changes. This study shows that retrieval of control-states is not limited to a specific task or context feature and therefore generalizes the notion that abstract control parameters are stored into trial-specific event files.

Open set task augmentation facilitates generalization of deep neural networks trained on small data sets

Many application scenarios for image recognition require learning of deep networks from small sample sizes in the order of a few hundred samples per class. Then, avoiding overfitting is critical. Common techniques to address overfitting are transfer learning, reduction of model complexity and artificial enrichment of the available data by, e.g., data augmentation. A key idea proposed in this paper is to incorporate additional samples into the training that do not belong to the classes of the target task. This can be accomplished by formulating the original classification task as an open set classification task. While the original closed set classification task is not altered at inference time, the recast as open set classification task enables the inclusion of additional data during training. Hence, the original closed set classification task is augmented with an open set task during training. We therefore call the proposed approach open set task augmentation. In order to integrate additional task-unrelated samples into the training, we employ the entropic open set loss originally proposed for open set classification tasks and also show that similar results can be obtained with a modified sum of squared errors loss function. Learning with the proposed approach benefits from the integration of additional “unknown” samples, which are often available, e.g., from open data sets, and can then be easily integrated into the learning process. We show that this open set task augmentation can improve model performance even when these additional samples are rather few or far from the domain of the target task. The proposed approach is demonstrated on two exemplary scenarios based on subsets of the ImageNet and Food-101 data sets as well as with several network architectures and two loss functions. We further shed light on the impact of the entropic open set loss on the internal representations formed by the networks. Open set task augmentation is particularly valuable when no additional data from the target classes are available—a scenario often faced in practice.

Excluding Parallel Execution to Improve Global Fixed Priority Response Time Analysis

Response Time Analysis (RTA) is an effective method for testing the schedulability of real-time tasks on multiprocessor platforms. Existing RTAs for global fixed priority scheduling calculate the upper bound of the worst case response time of each task. Given a target task, existing RTAs first calculate the workload upper bound of each higher priority task (than the target task), and then calculate the interference on the target task by each higher priority task according to the obtained workload upper bounds. The workload of a task consists of three parts: carry-in, body and carry-out. The interference from all these three parts may be overestimated in existing RTAs. However, although the overestimation of the interference from body is the major factor that causes the low accuracy of existing RTAs, all existing work only focuses on how to reduce the overestimation of the interference from carry-in, and there is no method to reduce the overestimation of the interference from body or carry-out. In this work, we propose a method to calculate the lower bound of the accumulative time in which the target task and higher priority tasks are executed in parallel. By excluding the parallel execution time from the interference, we derive a new RTA test that can reduce the overestimation of the interference from all three parts of the workload. Extensive experiments are conducted to verify the superior performance of the proposed RTA test.

Teaching Multiple Inverse Reinforcement Learners

In this paper, we propose the first machine teaching algorithm for multiple inverse reinforcement learners. As our initial contribution, we formalize the problem of optimally teaching a sequential task to a heterogeneous class of learners. We then contribute a theoretical analysis of such problem, identifying conditions under which it is possible to conduct such teaching using the same demonstration for all learners. Our analysis shows that, contrary to other teaching problems, teaching a sequential task to a heterogeneous class of learners with a single demonstration may not be possible, as the differences between individual agents increase. We then contribute two algorithms that address the main difficulties identified by our theoretical analysis. The first algorithm, which we dub SplitTeach, starts by teaching the class as a whole until all students have learned all that they can learn as a group; it then teaches each student individually, ensuring that all students are able to perfectly acquire the target task. The second approach, which we dub JointTeach, selects a single demonstration to be provided to the whole class so that all students learn the target task as well as a single demonstration allows. While SplitTeach ensures optimal teaching at the cost of a bigger teaching effort, JointTeach ensures minimal effort, although the learners are not guaranteed to perfectly recover the target task. We conclude by illustrating our methods in several simulation domains. The simulation results agree with our theoretical findings, showcasing that indeed class teaching is not possible in the presence of heterogeneous students. At the same time, they also illustrate the main properties of our proposed algorithms: in all domains, SplitTeach guarantees perfect teaching and, in terms of teaching effort, is always at least as good as individualized teaching (often better); on the other hand, JointTeach attains minimal teaching effort in all domains, even if sometimes it compromises the teaching performance.

Towards Robust Model Reuse in the Presence of Latent Domains

Model reuse tries to adapt well pre-trained models to a new target task, without access of raw data. It attracts much attention since it reduces the learning resources. Previous model reuse studies typically operate in a single-domain scenario, i.e., the target samples arise from one single domain. However, in practice the target samples often arise from multiple latent or unknown domains, e.g., the images for cars may arise from latent domains such as photo, line drawing, cartoon, etc. The methods based on single-domain may no longer be feasible for multiple latent domains and may sometimes even lead to performance degeneration. To address the above issue, in this paper we propose the MRL (Model Reuse for multiple Latent domains) method. Both domain characteristics and pre-trained models are considered for the exploration of instances in the target task. Theoretically, the overall considerations are packed in a bi-level optimization framework with a reliable generalization. Moreover, through an ensemble of multiple models, the model robustness is improved with a theoretical guarantee. Empirical results on diverse real-world data sets clearly validate the effectiveness of proposed algorithms.

Improving Heterogeneous Network Knowledge Transfer Based on the Principle of Generative Adversarial

Deep learning requires a large amount of datasets to train deep neural network models for specific tasks, and thus training of a new model is a very costly task. Research on transfer networks used to reduce training costs will be the next turning point in deep learning research. The use of source task models to help reduce the training costs of the target task models, especially heterogeneous systems, is a problem we are studying. In order to quickly obtain an excellent target task model driven by the source task model, we propose a novel transfer learning approach. The model linearly transforms the feature mapping of the target domain and increases the weight value for feature matching to realize the knowledge transfer between heterogeneous networks and add a domain discriminator based on the principle of generative adversarial to speed up feature mapping and learning. Most importantly, this paper proposes a new objective function optimization scheme to complete the model training. It successfully combines the generative adversarial network with the weight feature matching method to ensure that the target model learns the most beneficial features from the source domain for its task. Compared with the previous transfer algorithm, our training results are excellent under the same benchmark for image recognition tasks.

Auto-sharing parameters for transfer learning based on multi-objective optimization

Transfer learning methods exploit similarities between different datasets to improve the performance of the target task by transferring knowledge from source tasks to the target task. “What to transfer” is a main research issue in transfer learning. The existing transfer learning method generally needs to acquire the shared parameters by integrating human knowledge. However, in many real applications, an understanding of which parameters can be shared is unknown beforehand. Transfer learning model is essentially a special multi-objective optimization problem. Consequently, this paper proposes a novel auto-sharing parameter technique for transfer learning based on multi-objective optimization and solves the optimization problem by using a multi-swarm particle swarm optimizer. Each task objective is simultaneously optimized by a sub-swarm. The current best particle from the sub-swarm of the target task is used to guide the search of particles of the source tasks and vice versa. The target task and source task are jointly solved by sharing the information of the best particle, which works as an inductive bias. Experiments are carried out to evaluate the proposed algorithm on several synthetic data sets and two real-world data sets of a school data set and a landmine data set, which show that the proposed algorithm is effective.

Continuous Classification of Locomotion in Response to Task Complexity and Anticipatory State

ObjectiveIntent recognition in lower-extremity assistive devices (e.g., prostheses and exoskeletons) is typically limited to either recognition of steady-state locomotion or changes of terrain (e.g., level ground to stair) occurring in a straight-line path and under anticipated condition. Stability is highly affected during non-steady changes of direction such as cuts especially when they are unanticipated, posing high risk of fall-related injuries. Here, we studied the influence of changes of direction and user anticipation on task recognition, and accordingly introduced classification schemes accommodating such effects.MethodsA linear discriminant analysis (LDA) classifier continuously classified straight-line walking, sidestep/crossover cuts (single transitions), and cuts-to-stair locomotion (mixed transitions) performed under varied task anticipatory conditions. Training paradigms with varying levels of anticipated/unanticipated exposures and analysis windows of size 100–600 ms were examined.ResultsMore accurate classification of anticipated relative to unanticipated tasks was observed. Including bouts of target task in the training data was necessary to improve generalization to unanticipated locomotion. Only up to two bouts of target task were sufficient to reduce errors to <20% in unanticipated mixed transitions, whereas, in single transitions and straight walking, substantial unanticipated information (i.e., five bouts) was necessary to achieve similar outcomes. Window size modifications did not have a significant influence on classification performance.ConclusionAdjusting the training paradigm helps to achieve classification schemes capable of adapting to changes of direction and task anticipatory state.SignificanceThe findings could provide insight into developing classification schemes that can adapt to changes of direction and user anticipation. They could inform intent recognition strategies for controlling lower-limb assistive to robustly handle “unknown” circumstances, and thus deliver increased level of reliability and safety.

TransBERT

Recent advances, such as GPT, BERT, and RoBERTa, have shown success in incorporating a pre-trained transformer language model and fine-tuning operations to improve downstream NLP systems. However, this framework still has some fundamental problems in effectively incorporating supervised knowledge from other related tasks. In this study, we investigate a transferable BERT (TransBERT) training framework, which can transfer not only general language knowledge from large-scale unlabeled data but also specific kinds of knowledge from various semantically related supervised tasks, for a target task. Particularly, we propose utilizing three kinds of transfer tasks, including natural language inference, sentiment classification, and next action prediction, to further train BERT based on a pre-trained model. This enables the model to get a better initialization for the target task. We take story-ending prediction as the target task to conduct experiments. The final results of 96.0% and 95.0% accuracy on two versions of Story Cloze Test datasets dramatically outperform previous state-of-the-art baseline methods. Several comparative experiments give some helpful suggestions on how to select transfer tasks to improve BERT. Furthermore, experiments on six English and three Chinese datasets show that TransBERT generalizes well to other tasks, languages, and pre-trained models.

Phase Transitions in Transfer Learning for High-Dimensional Perceptrons

Transfer learning seeks to improve the generalization performance of a target task by exploiting the knowledge learned from a related source task. Central questions include deciding what information one should transfer and when transfer can be beneficial. The latter question is related to the so-called negative transfer phenomenon, where the transferred source information actually reduces the generalization performance of the target task. This happens when the two tasks are sufficiently dissimilar. In this paper, we present a theoretical analysis of transfer learning by studying a pair of related perceptron learning tasks. Despite the simplicity of our model, it reproduces several key phenomena observed in practice. Specifically, our asymptotic analysis reveals a phase transition from negative transfer to positive transfer as the similarity of the two tasks moves past a well-defined threshold.