Search

Large Language Models (LLMs) are regularly updated to enhance performance, typically through changes in data or architecture. Within the update process, developers often prioritize improving overall performance metrics, paying less attention to maintaining compatibility with earlier model versions. Instance-level degradation (instance regression) of performance from one model version to the next can interfere with a user's mental model of the capabilities of a particular language model. Users having to adapt their mental model with every update can lead to dissatisfaction, especially when the new model has degraded compared to a prior version for a known use case (model update regression). We find that when pretrained LLM base models are updated, fine-tuned user-facing downstream task adapters experience negative flips -- previously correct instances are now predicted incorrectly. We observe model update regression between different model versions on a diverse set of tasks and models, even when the downstream task training procedures remain identical. We argue for the importance of maintaining model update compatibility during updates, and present evaluation metrics designed specifically for generative tasks, while also being applicable to discriminative tasks. We propose a training strategy to minimize the extent of instance regression in model updates, involving training of a compatibility adapter that can enhance task fine-tuned language models. We show negative flips reduce by up to 40% e.g. when updating Llama 1 to Llama 2 with our proposed method.

AI methods are finding an increasing number of applications, but their often black-box nature has raised concerns about accountability and trust. The field of explainable artificial intelligence (XAI) has emerged in response to the need for human understanding of AI models. Evolutionary computation (EC), as a family of powerful optimization and learning tools, has significant potential to contribute to XAI. In this paper, we provide an introduction to XAI and review various techniques in current use for explaining machine learning (ML) models. We then focus on how EC can be used in XAI, and review some XAI approaches which incorporate EC techniques. Additionally, we discuss the application of XAI principles within EC itself, examining how these principles can shed some light on the behavior and outcomes of EC algorithms in general, on the (automatic) configuration of these algorithms, and on the underlying problem landscapes that these algorithms optimize. Finally, we discuss some open challenges in XAI and opportunities for future research in this field using EC. Our aim is to demonstrate that EC is well-suited for addressing current problems in explainability and to encourage further exploration of these methods to contribute to the development of more transparent and trustworthy ML models and EC algorithms., Comment: 29 pages, 4 figures. arXiv admin note: substantial text overlap with arXiv:2306.14786

Skeleton-based action representation learning aims to interpret and understand human behaviors by encoding the skeleton sequences, which can be categorized into two primary training paradigms: supervised learning and self-supervised learning. However, the former one-hot classification requires labor-intensive predefined action categories annotations, while the latter involves skeleton transformations (e.g., cropping) in the pretext tasks that may impair the skeleton structure. To address these challenges, we introduce a novel skeleton-based training framework (C$^2$VL) based on Cross-modal Contrastive learning that uses the progressive distillation to learn task-agnostic human skeleton action representation from the Vision-Language knowledge prompts. Specifically, we establish the vision-language action concept space through vision-language knowledge prompts generated by pre-trained large multimodal models (LMMs), which enrich the fine-grained details that the skeleton action space lacks. Moreover, we propose the intra-modal self-similarity and inter-modal cross-consistency softened targets in the cross-modal representation learning process to progressively control and guide the degree of pulling vision-language knowledge prompts and corresponding skeletons closer. These soft instance discrimination and self-knowledge distillation strategies contribute to the learning of better skeleton-based action representations from the noisy skeleton-vision-language pairs. During the inference phase, our method requires only the skeleton data as the input for action recognition and no longer for vision-language prompts. Extensive experiments on NTU RGB+D 60, NTU RGB+D 120, and PKU-MMD datasets demonstrate that our method outperforms the previous methods and achieves state-of-the-art results. Code is available at: https://github.com/cseeyangchen/C2VL., Comment: Accepted by IEEE Transactions on Multimedia

Scandium-doped aluminum nitride (AlScN) with an asymmetric hexagonal wurtzite structure exhibits enhanced second-order nonlinear and piezoelectric properties compared to aluminum nitride (AlN), while maintaining a relatively large bandgap. It provides a promising platform for photonic integration and facilitates the seamless integration of passive and active functional devices. Here, we present the design, fabrication, and characterization of AlScN EO micro-ring modulators, introducing active functionalities to the chip-scale AlScN platform. These waveguide-integrated EO modulators employ sputtered AlScN thin films as the light-guiding medium, and the entire fabrication process is compatible with complementary metal oxide semiconductor (CMOS) technology. We characterize the high-frequency performance of an AlScN modulator for the first time, extracting a maximum in-device effective EO coefficient of 2.86 pm/V at 12 GHz. The devices show a minimum half-wave voltage-length product of 3.12 V*cm and a 3-dB modulation bandwidth of approximately 22 GHz. Our work provides a promising modulation scheme for cost-effective silicon-integrated photonics systems.

Symmetry engineering is explicitly effective to manipulate and even create phases and orderings in strongly correlated materials. Flexural stress is universally practical to break the space-inversion or time-reversal symmetry. Here, by introducing strain gradient in a centrosymmetric antiferromagnet Sr2IrO4, the space-inversion symmetry is broken accompanying a non-equivalent O p-Ir d orbital hybridization along z axis. Thus, emergent polar phase and out-of-plane magnetic moment have been simultaneously observed in these asymmetric Sr2IrO4 thin films, which both are absent in its ground state. Furthermore, upon the application of magnetic field, such polarization can be controlled by modifying the occupied d orbitals through spin-orbit interaction, giving rise to a flexomagnetoelectric effect. This work provides a general strategy to artificially design multiple symmetries and ferroic orderings in strongly correlated systems.

While Automatic Speech Recognition (ASR) systems are widely used in many real-world applications, they often do not generalize well to new domains and need to be finetuned on data from these domains. However, target-domain data usually are not readily available in many scenarios. In this paper, we propose a new strategy for adapting ASR models to new target domains without any text or speech from those domains. To accomplish this, we propose a novel data synthesis pipeline that uses a Large Language Model (LLM) to generate a target domain text corpus, and a state-of-the-art controllable speech synthesis model to generate the corresponding speech. We propose a simple yet effective in-context instruction finetuning strategy to increase the effectiveness of LLM in generating text corpora for new domains. Experiments on the SLURP dataset show that the proposed method achieves an average relative word error rate improvement of $28\%$ on unseen target domains without any performance drop in source domains.

The increasingly Large Language Models (LLMs) demonstrate stronger language understanding and generation capabilities, while the memory demand and computation cost of fine-tuning LLMs on downstream tasks are non-negligible. Besides, fine-tuning generally requires a certain amount of data from individual tasks whilst data collection cost is another issue to consider in real-world applications. In this work, we focus on Parameter-Efficient Fine-Tuning (PEFT) methods for few-shot Natural Language Generation (NLG), which freeze most parameters in LLMs and tune a small subset of parameters in few-shot cases so that memory footprint, training cost, and labeling cost are reduced while maintaining or even improving the performance. We propose a Scaled Prompt-Tuning (SPT) method which surpasses conventional PT with better performance and generalization ability but without an obvious increase in training cost. Further study on intermediate SPT suggests the superior transferability of SPT in few-shot scenarios, providing a recipe for data-deficient and computation-limited circumstances. Moreover, a comprehensive comparison of existing PEFT methods reveals that certain approaches exhibiting decent performance with modest training cost such as Prefix-Tuning in prior study could struggle in few-shot NLG tasks, especially on challenging datasets.

Data complexity analysis quantifies the hardness of constructing a predictive model on a given dataset. However, the effectiveness of existing data complexity measures can be challenged by the existence of irrelevant features and feature interactions in biological micro-array data. We propose a novel data complexity measure, depth, that leverages an evolutionary inspired feature selection algorithm to quantify the complexity of micro-array data. By examining feature subsets of varying sizes, the approach offers a novel perspective on data complexity analysis. Unlike traditional metrics, depth is robust to irrelevant features and effectively captures complexity stemming from feature interactions. On synthetic micro-array data, depth outperforms existing methods in robustness to irrelevant features and identifying complexity from feature interactions. Applied to case-control genotype and gene-expression micro-array datasets, the results reveal that a single feature of gene-expression data can account for over 90% of the performance of multi-feature model, confirming the adequacy of the commonly used differentially expressed gene (DEG) feature selection method for the gene expression data. Our study also demonstrates that constructing predictive models for genotype data is harder than gene expression data. The results in this paper provide evidence for the use of interpretable machine learning algorithms on microarray data.

Through genome-wide association studies (GWAS), disease susceptible genetic variables can be identified by comparing the genetic data of individuals with and without a specific disease. However, the discovery of these associations poses a significant challenge due to genetic heterogeneity and feature interactions. Genetic variables intertwined with these effects often exhibit lower effect-size, and thus can be difficult to be detected using machine learning feature selection methods. To address these challenges, this paper introduces a novel feature selection mechanism for GWAS, named Feature Co-selection Network (FCSNet). FCS-Net is designed to extract heterogeneous subsets of genetic variables from a network constructed from multiple independent feature selection runs based on a genetic algorithm (GA), an evolutionary learning algorithm. We employ a non-linear machine learning algorithm to detect feature interaction. We introduce the Community Risk Score (CRS), a synthetic feature designed to quantify the collective disease association of each variable subset. Our experiment showcases the effectiveness of the utilized GA-based feature selection method in identifying feature interactions through synthetic data analysis. Furthermore, we apply our novel approach to a case-control colorectal cancer GWAS dataset. The resulting synthetic features are then used to explain the genetic heterogeneity in an additional case-only GWAS dataset.

Thin-film lithium niobate (TFLN) based frequency doublers have been widely recognized as essential components for both classical and quantum optical communications. Nonetheless, the efficiency of these devices is hindered by imperfections present in the quasi-phase matching (QPM) spectrum. In this study, we present a thorough analysis of the spectral imperfections in TFLN frequency doublers with varying lengths, ranging from 5 mm to 15 mm. Employing a non-destructive diagnostic method based on scattered light imaging, we identify the sources and waveguide sections that contribute to the imperfections in the QPM spectrum. Furthermore, by mapping the TFLN film thickness across the entire waveguiding regions, we successfully reproduce the QPM spectra numerically, thus confirming the prominent influence of film thickness variations on the observed spectral imperfections. This comprehensive investigation provides valuable insights into the identification and mitigation of spectral imperfections in TFLN-based frequency doublers, paving the way toward the realization of nonlinear optical devices with enhanced efficiency and improved spectral fidelity.

Machine learning models are increasingly being used in critical sectors, but their black-box nature has raised concerns about accountability and trust. The field of explainable artificial intelligence (XAI) or explainable machine learning (XML) has emerged in response to the need for human understanding of these models. Evolutionary computing, as a family of powerful optimization and learning tools, has significant potential to contribute to XAI/XML. In this chapter, we provide a brief introduction to XAI/XML and review various techniques in current use for explaining machine learning models. We then focus on how evolutionary computing can be used in XAI/XML, and review some approaches which incorporate EC techniques. We also discuss some open challenges in XAI/XML and opportunities for future research in this field using EC. Our aim is to demonstrate that evolutionary computing is well-suited for addressing current problems in explainability, and to encourage further exploration of these methods to contribute to the development of more transparent, trustworthy and accountable machine learning models.

Optical neural networks (ONNs) enable high speed parallel and energy efficient processing compared to conventional digital electronic counterparts. However, realizing large scale systems is an open problem. Among various integrated and non-integrated ONNs, free-space diffractive ONNs benefit from a large number of pixels of spatial light modulators to realize millions of neurons. However, a significant fraction of computation time and energy is consumed by the nonlinear activation function that is typically implemented using a camera sensor. Here, we propose a novel surface-normal photodetector (SNPD) with a nonlinear response to replace the camera sensor that enables about three orders of magnitude faster (5.7 us response time) and more energy efficient (less than 10 nW/pixel) response. Direct efficient vertical optical coupling, polarization insensitivity, inherent nonlinearity with no control electronics, low optical power requirements, and the possibility of implementing large scale arrays make the SNPD a promising nonlinear activation function for diffractive ONNs. To show the applicability, successful classification simulation of MNIST and Fashion MNIST datasets using the measured response of SNPD with accuracy comparable to that of an ideal ReLU function are demonstrated.

Adapting generic speech recognition models to specific individuals is a challenging problem due to the scarcity of personalized data. Recent works have proposed boosting the amount of training data using personalized text-to-speech synthesis. Here, we ask two fundamental questions about this strategy: when is synthetic data effective for personalization, and why is it effective in those cases? To address the first question, we adapt a state-of-the-art automatic speech recognition (ASR) model to target speakers from four benchmark datasets representative of different speaker types. We show that ASR personalization with synthetic data is effective in all cases, but particularly when (i) the target speaker is underrepresented in the global data, and (ii) the capacity of the global model is limited. To address the second question of why personalized synthetic data is effective, we use controllable speech synthesis to generate speech with varied styles and content. Surprisingly, we find that the text content of the synthetic data, rather than style, is important for speaker adaptation. These results lead us to propose a data selection strategy for ASR personalization based on speech content., Comment: ICASSP 2023

Genotype-to-phenotype mappings translate genotypic variations such as mutations into phenotypic changes. Neutrality is the observation that some mutations do not lead to phenotypic changes. Studying the search trajectories in genotypic and phenotypic spaces, especially through neutral mutations, helps us to better understand the progression of evolution and its algorithmic behaviour. In this study, we visualise the search trajectories of a genetic programming system as graph-based models, where nodes are genotypes/phenotypes and edges represent their mutational transitions. We also quantitatively measure the characteristics of phenotypes including their genotypic abundance (the requirement for neutrality) and Kolmogorov complexity. We connect these quantified metrics with search trajectory visualisations, and find that more complex phenotypes are under-represented by fewer genotypes and are harder for evolution to discover. Less complex phenotypes, on the other hand, are over-represented by genotypes, are easier to find, and frequently serve as stepping-stones for evolution.

Transformer-based architectures like BERT have achieved great success in a wide range of Natural Language tasks. Despite their decent performance, the models still have numerous parameters and high computational complexity, impeding their deployment in resource-constrained environments. Post-Training Quantization (PTQ), which enables low-bit computations without extra training, could be a promising tool. In this work, we conduct an empirical evaluation of three PTQ methods on BERT-Base and BERT-Large: Linear Quantization (LQ), Analytical Clipping for Integer Quantization (ACIQ), and Outlier Channel Splitting (OCS). OCS theoretically surpasses the others in minimizing the Mean Square quantization Error and avoiding distorting the weights' outliers. That is consistent with the evaluation results of most language tasks of GLUE benchmark and a reading comprehension task, SQuAD. Moreover, low-bit quantized BERT models could outperform the corresponding 32-bit baselines on several small language tasks, which we attribute to the alleviation of over-parameterization. We further explore the limit of quantization bit and show that OCS could quantize BERT-Base and BERT-Large to 3-bits and retain 98% and 96% of the performance on the GLUE benchmark accordingly. Moreover, we conduct quantization on the whole BERT family, i.e., BERT models in different configurations, and comprehensively evaluate their performance on the GLUE benchmark and SQuAD, hoping to provide valuable guidelines for their deployment in various computation environments.

This paper explores the possibility of using visual object detection techniques for word localization in speech data. Object detection has been thoroughly studied in the contemporary literature for visual data. Noting that an audio can be interpreted as a 1-dimensional image, object localization techniques can be fundamentally useful for word localization. Building upon this idea, we propose a lightweight solution for word detection and localization. We use bounding box regression for word localization, which enables our model to detect the occurrence, offset, and duration of keywords in a given audio stream. We experiment with LibriSpeech and train a model to localize 1000 words. Compared to existing work, our method reduces model size by 94%, and improves the F1 score by 6.5\%.

Recent studies on Learning to Optimize (L2O) suggest a promising path to automating and accelerating the optimization procedure for complicated tasks. Existing L2O models parameterize optimization rules by neural networks, and learn those numerical rules via meta-training. However, they face two common pitfalls: (1) scalability: the numerical rules represented by neural networks create extra memory overhead for applying L2O models, and limit their applicability to optimizing larger tasks; (2) interpretability: it is unclear what an L2O model has learned in its black-box optimization rule, nor is it straightforward to compare different L2O models in an explainable way. To avoid both pitfalls, this paper proves the concept that we can "kill two birds by one stone", by introducing the powerful tool of symbolic regression to L2O. In this paper, we establish a holistic symbolic representation and analysis framework for L2O, which yields a series of insights for learnable optimizers. Leveraging our findings, we further propose a lightweight L2O model that can be meta-trained on large-scale problems and outperformed human-designed and tuned optimizers. Our work is set to supply a brand-new perspective to L2O research. Codes are available at: https://github.com/VITA-Group/Symbolic-Learning-To-Optimize., Comment: Published as a conference paper at ICLR 2022

In hyperspectral, high-quality spectral signals convey subtle spectral differences to distinguish similar materials, thereby providing unique advantage for anomaly detection. Hence fine spectra of anomalous pixels can be effectively screened out from heterogeneous background pixels. Since the same materials have similar characteristics in spatial and spectral dimension, detection performance can be significantly enhanced by jointing spatial and spectral information. In this paper, a spectralspatial fusion anomaly detection (SSFAD) method is proposed for hyperspectral imagery. First, original spectral signals are mapped to a local linear background space composed of median and mean with high confidence, where saliency weight and feature enhancement strategies are implemented to obtain an initial detection map in spectral domain. Futhermore, to make full use of similarity information of local background around testing pixel, a new detector is designed to extract the local similarity spatial features of patch images in spatial domain. Finally, anomalies are detected by adaptively combining the spectral and spatial detection maps. The experimental results demonstrate that our proposed method has superior detection performance than traditional methods.