Transfer learning is known to perform efficiently in many applications empirically, yet limited literature reports the mechanism behind the scene. This study establishes both formal derivations and heuristic analysis to formulate the theory of transfer learning in deep learning. Our framework utilizing layer variational analysis proves that the success of transfer learning can be guaranteed with corresponding data conditions. Moreover, our theoretical calculation yields intuitive interpretations towards the knowledge transfer process. Subsequently, an alternative method for network-based transfer learning is derived. The method shows an increase in efficiency and accuracy for domain adaptation. It is particularly advantageous when new domain data is sufficiently sparse during adaptation. Numerical experiments over diverse tasks validated our theory and verified that our analytic expression achieved better performance in domain adaptation than the gradient descent method.
Purpose: In curriculum learning, the idea is to train on easier samples first and gradually increase the difficulty, while in self-paced learning, a pacing function defines the speed to adapt the training progress. While both methods heavily rely on the ability to score the difficulty of data samples, an optimal scoring function is still under exploration. Methodology: Distillation is a knowledge transfer approach where a teacher network guides a student network by feeding a sequence of random samples. We argue that guiding student networks with an efficient curriculum strategy can improve model generalization and robustness. For this purpose, we design an uncertainty-based paced curriculum learning in self distillation for medical image segmentation. We fuse the prediction uncertainty and annotation boundary uncertainty to develop a novel paced-curriculum distillation (PCD). We utilize the teacher model to obtain prediction uncertainty and spatially varying label smoothing with Gaussian kernel to generate segmentation boundary uncertainty from the annotation. We also investigate the robustness of our method by applying various types and severity of image perturbation and corruption. Results: The proposed technique is validated on two medical datasets of breast ultrasound image segmentation and robotassisted surgical scene segmentation and achieved significantly better performance in terms of segmentation and robustness. Conclusion: P-CD improves the performance and obtains better generalization and robustness over the dataset shift. While curriculum learning requires extensive tuning of hyper-parameters for pacing function, the level of performance improvement suppresses this limitation.
We consider gradient-based optimisation of wide, shallow neural networks, where the output of each hidden node is scaled by a positive parameter. The scaling parameters are non-identical, differing from the classical Neural Tangent Kernel (NTK) parameterisation. We prove that for large such neural networks, with high probability, gradient flow and gradient descent converge to a global minimum and can learn features in some sense, unlike in the NTK parameterisation. We perform experiments illustrating our theoretical results and discuss the benefits of such scaling in terms of prunability and transfer learning.
Few-shot classification consists of a training phase where a model is learned on a relatively large dataset and an adaptation phase where the learned model is adapted to previously-unseen tasks with limited labeled samples. In this paper, we empirically prove that the training algorithm and the adaptation algorithm can be completely disentangled, which allows algorithm analysis and design to be done individually for each phase. Our meta-analysis for each phase reveals several interesting insights that may help better understand key aspects of few-shot classification and connections with other fields such as visual representation learning and transfer learning. We hope the insights and research challenges revealed in this paper can inspire future work in related directions. Code and pre-trained models (in PyTorch) are available at https://github.com/Frankluox/CloserLookAgainFewShot.
Accurate modeling of student knowledge is essential for large-scale online learning systems that are increasingly used for student training. Knowledge tracing aims to model student knowledge state given the student's sequence of learning activities. Modern Knowledge tracing (KT) is usually formulated as a supervised sequence learning problem to predict students' future practice performance according to their past observed practice scores by summarizing student knowledge state as a set of evolving hidden variables. Because of this formulation, many current KT solutions are not fit for modeling student learning from non-assessed learning activities with no explicit feedback or score observation (e.g., watching video lectures that are not graded). Additionally, these models cannot explicitly represent the dynamics of knowledge transfer among different learning activities, particularly between the assessed (e.g., quizzes) and non-assessed (e.g., video lectures) learning activities. In this paper, we propose Transition-Aware Multi-activity Knowledge Tracing (TAMKOT), which models knowledge transfer between learning materials, in addition to student knowledge, when students transition between and within assessed and non-assessed learning materials. TAMKOT is formulated as a deep recurrent multi-activity learning model that explicitly learns knowledge transfer by activating and learning a set of knowledge transfer matrices, one for each transition type between student activities. Accordingly, our model allows for representing each material type in a different yet transferrable latent space while maintaining student knowledge in a shared space. We evaluate our model on three real-world publicly available datasets and demonstrate TAMKOT's capability in predicting student performance and modeling knowledge transfer.
Large-scale pre-trained language models (PLMs) have shown great potential in natural language processing tasks. Leveraging the capabilities of PLMs to enhance automatic speech recognition (ASR) systems has also emerged as a promising research direction. However, previous works may be limited by the inflexible structures of PLMs and the insufficient utilization of PLMs. To alleviate these problems, we propose the hierarchical knowledge distillation (HKD) on the continuous integrate-and-fire (CIF) based ASR models. To transfer knowledge from PLMs to the ASR models, HKD employs cross-modal knowledge distillation with contrastive loss at the acoustic level and knowledge distillation with regression loss at the linguistic level. Compared with the original CIF-based model, our method achieves 15% and 9% relative error rate reduction on the AISHELL-1 and LibriSpeech datasets, respectively.
Bayesian model comparison (BMC) offers a principled approach for assessing the relative merits of competing computational models and propagating uncertainty into model selection decisions. However, BMC is often intractable for the popular class of hierarchical models due to their high-dimensional nested parameter structure. To address this intractability, we propose a deep learning method for performing BMC on any set of hierarchical models which can be instantiated as probabilistic programs. Since our method enables amortized inference, it allows efficient re-estimation of posterior model probabilities and fast performance validation prior to any real-data application. In a series of extensive validation studies, we benchmark the performance of our method against the state-of-the-art bridge sampling method and demonstrate excellent amortized inference across all BMC settings. We then showcase our method by comparing four hierarchical evidence accumulation models that have previously been deemed intractable for BMC due to partly implicit likelihoods. Additionally, we demonstrate how transfer learning can be leveraged to enhance training efficiency. We provide reproducible code for all analyses and an open-source implementation of our method.
The generation of molecules with desired properties has become increasingly popular, revolutionizing the way scientists design molecular structures and providing valuable support for chemical and drug design. However, despite the potential of language models in molecule generation, they face challenges such as generating syntactically or chemically flawed molecules, having narrow domain focus, and struggling to create diverse and feasible molecules due to limited annotated data or external molecular databases. To tackle these challenges, we introduce MolGen, a pre-trained molecular language model tailored specifically for molecule generation. Through the reconstruction of over 100 million molecular SELFIES, MolGen internalizes structural and grammatical insights. This is further enhanced by domain-agnostic molecular prefix tuning, fostering robust knowledge transfer across diverse domains. Importantly, our chemical feedback paradigm steers the model away from molecular hallucinations, ensuring alignment between the model's estimated probabilities and real-world chemical preferences. Extensive experiments on well-known benchmarks underscore MolGen's optimization capabilities in properties such as penalized logP, QED, and molecular docking. Additional analyses confirm its proficiency in accurately capturing molecule distributions, discerning intricate structural patterns, and efficiently exploring the chemical space. Code is available at https://github.com/zjunlp/MolGen.
The Segmented Planar Imaging Detector for Electro-Optical Reconnaissance (SPIDER) is an optical interferometric imaging device that aims to offer an alternative to the large space telescope designs of today with reduced size, weight and power consumption. This is achieved through interferometric imaging. State-of-the-art methods for reconstructing images from interferometric measurements adopt proximal optimization techniques, which are computationally expensive and require handcrafted priors. In this work we present two data-driven approaches for reconstructing images from measurements made by the SPIDER instrument. These approaches use deep learning to learn prior information from training data, increasing the reconstruction quality, and significantly reducing the computation time required to recover images by orders of magnitude. Reconstruction time is reduced to ${\sim} 10$ milliseconds, opening up the possibility of real-time imaging with SPIDER for the first time. Furthermore, we show that these methods can also be applied in domains where training data is scarce, such as astronomical imaging, by leveraging transfer learning from domains where plenty of training data are available.
Precise load forecasting in buildings could increase the bill savings potential and facilitate optimized strategies for power generation planning. With the rapid evolution of computer science, data-driven techniques, in particular the Deep Learning models, have become a promising solution for the load forecasting problem. These models have showed accurate forecasting results; however, they need abundance amount of historical data to maintain the performance. Considering the new buildings and buildings with low resolution measuring equipment, it is difficult to get enough historical data from them, leading to poor forecasting performance. In order to adapt Deep Learning models for buildings with limited and scarce data, this paper proposes a Building-to-Building Transfer Learning framework to overcome the problem and enhance the performance of Deep Learning models. The transfer learning approach was applied to a new technique known as Transformer model due to its efficacy in capturing data trends. The performance of the algorithm was tested on a large commercial building with limited data. The result showed that the proposed approach improved the forecasting accuracy by 56.8% compared to the case of conventional deep learning where training from scratch is used. The paper also compared the proposed Transformer model to other sequential deep learning models such as Long-short Term Memory (LSTM) and Recurrent Neural Network (RNN). The accuracy of the transformer model outperformed other models by reducing the root mean square error to 0.009, compared to LSTM with 0.011 and RNN with 0.051.
In this paper, a new perspective is suggested for unsupervised Ontology Matching (OM) or Ontology Alignment (OA) by treating it as a translation task. Ontologies are represented as graphs, and the translation is performed from a node in the source ontology graph to a path in the target ontology graph. The proposed framework, Truveta Mapper (TM), leverages a multi-task sequence-to-sequence transformer model to perform alignment across multiple ontologies in a zero-shot, unified and end-to-end manner. Multi-tasking enables the model to implicitly learn the relationship between different ontologies via transfer-learning without requiring any explicit cross-ontology manually labeled data. This also enables the formulated framework to outperform existing solutions for both runtime latency and alignment quality. The model is pre-trained and fine-tuned only on publicly available text corpus and inner-ontologies data. The proposed solution outperforms state-of-the-art approaches, Edit-Similarity, LogMap, AML, BERTMap, and the recently presented new OM frameworks in Ontology Alignment Evaluation Initiative (OAEI22), offers log-linear complexity, and overall makes the OM task efficient and more straightforward without much post-processing involving mapping extension or mapping repair. We are open sourcing our solution.
Several high-resource Text to Speech (TTS) systems currently produce natural, well-established human-like speech. In contrast, low-resource languages, including Arabic, have very limited TTS systems due to the lack of resources. We propose a fully unsupervised method for building TTS, including automatic data selection and pre-training/fine-tuning strategies for TTS training, using broadcast news as a case study. We show how careful selection of data, yet smaller amounts, can improve the efficiency of TTS system in generating more natural speech than a system trained on a bigger dataset. We adopt to propose different approaches for the: 1) data: we applied automatic annotations using DNSMOS, automatic vowelization, and automatic speech recognition (ASR) for fixing transcriptions' errors; 2) model: we used transfer learning from high-resource language in TTS model and fine-tuned it with one hour broadcast recording then we used this model to guide a FastSpeech2-based Conformer model for duration. Our objective evaluation shows 3.9% character error rate (CER), while the groundtruth has 1.3% CER. As for the subjective evaluation, where 1 is bad and 5 is excellent, our FastSpeech2-based Conformer model achieved a mean opinion score (MOS) of 4.4 for intelligibility and 4.2 for naturalness, where many annotators recognized the voice of the broadcaster, which proves the effectiveness of our proposed unsupervised method.
In natural language processing (NLP), code-mixing (CM) is a challenging task, especially when the mixed languages include dialects. In Southeast Asian countries such as Singapore, Indonesia, and Malaysia, Hokkien-Mandarin is the most widespread code-mixed language pair among Chinese immigrants, and it is also common in Taiwan. However, dialects such as Hokkien often have a scarcity of resources and the lack of an official writing system, limiting the development of dialect CM research. In this paper, we propose a method to construct a Hokkien-Mandarin CM dataset to mitigate the limitation, overcome the morphological issue under the Sino-Tibetan language family, and offer an efficient Hokkien word segmentation method through a linguistics-based toolkit. Furthermore, we use our proposed dataset and employ transfer learning to train the XLM (cross-lingual language model) for translation tasks. To fit the code-mixing scenario, we adapt XLM slightly. We found that by using linguistic knowledge, rules, and language tags, the model produces good results on CM data translation while maintaining monolingual translation quality.
State-of-the-art video-text retrieval (VTR) methods typically involve fully fine-tuning a pre-trained model (e.g. CLIP) on specific datasets. However, this can result in significant storage costs in practical applications as a separate model per task must be stored. To address this issue, we present our pioneering work that enables parameter-efficient VTR using a pre-trained model, with only a small number of tunable parameters during training. Towards this goal, we propose a new method dubbed Multimodal Video Adapter (MV-Adapter) for efficiently transferring the knowledge in the pre-trained CLIP from image-text to video-text. Specifically, MV-Adapter utilizes bottleneck structures in both video and text branches, along with two novel components. The first is a Temporal Adaptation Module that is incorporated in the video branch to introduce global and local temporal contexts. We also train weights calibrations to adjust to dynamic variations across frames. The second is Cross Modality Tying that generates weights for video/text branches through sharing cross modality factors, for better aligning between modalities. Thanks to above innovations, MV-Adapter can achieve comparable or better performance than standard full fine-tuning with negligible parameters overhead. Notably, MV-Adapter consistently outperforms various competing methods in V2T/T2V tasks with large margins on five widely used VTR benchmarks (MSR-VTT, MSVD, LSMDC, DiDemo, and ActivityNet).
In this paper, we propose to exploit the side-tuning framework for multimodal document classification. Side-tuning is a methodology for network adaptation recently introduced to solve some of the problems related to previous approaches. Thanks to this technique it is actually possible to overcome model rigidity and catastrophic forgetting of transfer learning by fine-tuning. The proposed solution uses off-the-shelf deep learning architectures leveraging the side-tuning framework to combine a base model with a tandem of two side networks. We show that side-tuning can be successfully employed also when different data sources are considered, e.g. text and images in document classification. The experimental results show that this approach pushes further the limit for document classification accuracy with respect to the state of the art.
Two-dimensional (2D) materials have wide applications in superconductors, quantum, and topological materials. However, their rational design is not well established, and currently less than 6,000 experimentally synthesized 2D materials have been reported. Recently, deep learning, data-mining, and density functional theory (DFT)-based high-throughput calculations are widely performed to discover potential new materials for diverse applications. Here we propose a generative material design pipeline, namely material transformer generator(MTG), for large-scale discovery of hypothetical 2D materials. We train two 2D materials composition generators using self-learning neural language models based on Transformers with and without transfer learning. The models are then used to generate a large number of candidate 2D compositions, which are fed to known 2D materials templates for crystal structure prediction. Next, we performed DFT computations to study their thermodynamic stability based on energy-above-hull and formation energy. We report four new DFT-verified stable 2D materials with zero e-above-hull energies, including NiCl$_4$, IrSBr, CuBr$_3$, and CoBrCl. Our work thus demonstrates the potential of our MTG generative materials design pipeline in the discovery of novel 2D materials and other functional materials.
Transfer learning is a promising method for AOI applications since it can significantly shorten sample collection time and improve efficiency in today's smart manufacturing. However, related research enhanced the network models by applying TL without considering the domain similarity among datasets, the data long-tailedness of a source dataset, and mainly used linear transformations to mitigate the lack of samples. This research applies model-based TL via domain similarity to improve the overall performance and data augmentation in both target and source domains to enrich the data quality and reduce the imbalance. Given a group of source datasets from similar industrial processes, we define which group is the most related to the target through the domain discrepancy score and the number of samples each has. Then, we transfer the chosen pre-trained backbone weights to train and fine-tune the target network. Our research suggests increases in the F1 score and the PR curve up to 20% compared with TL using benchmark datasets.
In-context learning (ICL) is a type of prompting where a transformer model operates on a sequence of (input, output) examples and performs inference on-the-fly. In this work, we formalize in-context learning as an algorithm learning problem where a transformer model implicitly constructs a hypothesis function at inference-time. We first explore the statistical aspects of this abstraction through the lens of multitask learning: We obtain generalization bounds for ICL when the input prompt is (1) a sequence of i.i.d. (input, label) pairs or (2) a trajectory arising from a dynamical system. The crux of our analysis is relating the excess risk to the stability of the algorithm implemented by the transformer. We characterize when transformer/attention architecture provably obeys the stability condition and also provide empirical verification. For generalization on unseen tasks, we identify an inductive bias phenomenon in which the transfer learning risk is governed by the task complexity and the number of MTL tasks in a highly predictable manner. Finally, we provide numerical evaluations that (1) demonstrate transformers can indeed implement near-optimal algorithms on classical regression problems with i.i.d. and dynamic data, (2) provide insights on stability, and (3) verify our theoretical predictions.
An individualized treatment regime (ITR) is a decision rule that assigns treatments based on patients' characteristics. The value function of an ITR is the expected outcome in a counterfactual world had this ITR been implemented. Recently, there has been increasing interest in combining heterogeneous data sources, such as leveraging the complementary features of randomized controlled trial (RCT) data and a large observational study (OS). Usually, a covariate shift exists between the source and target population, rendering the source-optimal ITR unnecessarily optimal for the target population. We present an efficient and robust transfer learning framework for estimating the optimal ITR with right-censored survival data that generalizes well to the target population. The value function accommodates a broad class of functionals of survival distributions, including survival probabilities and restrictive mean survival times (RMSTs). We propose a doubly robust estimator of the value function, and the optimal ITR is learned by maximizing the value function within a pre-specified class of ITRs. We establish the $N^{-1/3}$ rate of convergence for the estimated parameter indexing the optimal ITR, and show that the proposed optimal value estimator is consistent and asymptotically normal even with flexible machine learning methods for nuisance parameter estimation. We evaluate the empirical performance of the proposed method by simulation studies and a real data application of sodium bicarbonate therapy for patients with severe metabolic acidaemia in the intensive care unit (ICU), combining a RCT and an observational study with heterogeneity.
Standard fine-tuning of language models typically performs well on in-distribution data, but suffers with generalization to distribution shifts. In this work, we aim to improve the generalization of adapter-based cross-lingual task transfer where such cross-language distribution shifts are imminent. We investigate scheduled unfreezing algorithms -- originally proposed to mitigate catastrophic forgetting in transfer learning -- for fine-tuning task adapters. Our experiments show that scheduled unfreezing methods close the gap to full fine-tuning and achieve stronger cross-lingual transfer performance, suggesting that these methods can go beyond just mitigating catastrophic forgetting. Next, aiming to understand these empirical findings, we investigate the learning dynamics of scheduled unfreezing using Fisher Information. Our experiments reveal that scheduled unfreezing induces different learning dynamics compared to standard fine-tuning, and provide evidence that the dynamics of Fisher Information during training correlate with cross-lingual generalization performance. We additionally propose a general scheduled unfreezing algorithm that achieves an average of 2 points improvement over four datasets compared to standard fine-tuning and provides empirical evidence for a theory-based justification of the heuristic unfreezing schedule for adapter training.