We consider the problem of model selection in a high-dimensional sparse linear regression model under privacy constraints. We propose a differentially private (DP) best subset selection method with strong statistical utility properties by adopting the well-known exponential mechanism for selecting the best model. To achieve computational expediency, we propose an efficient Metropolis-Hastings algorithm and under certain regularity conditions, we establish that it enjoys polynomial mixing time to its stationary distribution. As a result, we also establish both approximate differential privacy and statistical utility for the estimates of the mixed Metropolis-Hastings chain. Finally, we perform some illustrative experiments on simulated data showing that our algorithm can quickly identify active features under reasonable privacy budget constraints.
Choosing an appropriate tokenization scheme is often a bottleneck in low-resource cross-lingual transfer. To understand the downstream implications of text representation choices, we perform a comparative analysis on language models having diverse text representation modalities including 2 segmentation-based models (\texttt{BERT}, \texttt{mBERT}), 1 image-based model (\texttt{PIXEL}), and 1 character-level model (\texttt{CANINE}). First, we propose a scoring Language Quotient (LQ) metric capable of providing a weighted representation of both zero-shot and few-shot evaluation combined. Utilizing this metric, we perform experiments comprising 19 source languages and 133 target languages on three tasks (POS tagging, Dependency parsing, and NER). Our analysis reveals that image-based models excel in cross-lingual transfer when languages are closely related and share visually similar scripts. However, for tasks biased toward word meaning (POS, NER), segmentation-based models prove to be superior. Furthermore, in dependency parsing tasks where word relationships play a crucial role, models with their character-level focus, outperform others. Finally, we propose a recommendation scheme based on our findings to guide model selection according to task and language requirements.
The reasoning capabilities of LLM (Large Language Model) are widely acknowledged in recent research, inspiring studies on tool learning and autonomous agents. LLM serves as the "brain" of the agent, orchestrating multiple tools for collaborative multi-step task solving. Unlike methods invoking tools like calculators or weather APIs for straightforward tasks, multi-modal agents excel by integrating diverse AI models for complex challenges. However, current multi-modal agents neglect the significance of model selection: they primarily focus on the planning and execution phases, and will only invoke predefined task-specific models for each subtask, making the execution fragile. Meanwhile, other traditional model selection methods are either incompatible with or suboptimal for the multi-modal agent scenarios, due to ignorance of dependencies among subtasks arising by multi-step reasoning. To this end, we identify the key challenges therein and propose the $\textit{M}^3$ framework as a plug-in with negligible runtime overhead at test-time. This framework improves model selection and bolsters the robustness of multi-modal agents in multi-step reasoning. In the absence of suitable benchmarks, we create MS-GQA, a new dataset specifically designed to investigate the model selection challenge in multi-modal agents. Our experiments reveal that our framework enables dynamic model selection, considering both user inputs and subtask dependencies, thereby robustifying the overall reasoning process. Our code and benchmark: https://github.com/LINs-lab/M3.
The dominance of proprietary LLMs has led to restricted access and raised information privacy concerns. High-performing open-source alternatives are crucial for information-sensitive and high-volume applications but often lag behind in performance. To address this gap, we propose (1) A untargeted variant of iterative self-critique and self-refinement devoid of external influence. (2) A novel ranking metric - Performance, Refinement, and Inference Cost Score (PeRFICS) - to find the optimal model for a given task considering refined performance and cost. Our experiments show that SoTA open source models of varying sizes from 7B - 65B, on average, improve 8.2% from their baseline performance. Strikingly, even models with extremely small memory footprints, such as Vicuna-7B, show a 11.74% improvement overall and up to a 25.39% improvement in high-creativity, open ended tasks on the Vicuna benchmark. Vicuna-13B takes it a step further and outperforms ChatGPT post-refinement. This work has profound implications for resource-constrained and information-sensitive environments seeking to leverage LLMs without incurring prohibitive costs, compromising on performance and privacy. The domain-agnostic self-refinement process coupled with our novel ranking metric facilitates informed decision-making in model selection, thereby reducing costs and democratizing access to high-performing language models, as evidenced by case studies.
We propose a distributional framework for benchmarking socio-technical risks of foundation models with quantified statistical significance. Our approach hinges on a new statistical relative testing based on first and second order stochastic dominance of real random variables. We show that the second order statistics in this test are linked to mean-risk models commonly used in econometrics and mathematical finance to balance risk and utility when choosing between alternatives. Using this framework, we formally develop a risk-aware approach for foundation model selection given guardrails quantified by specified metrics. Inspired by portfolio optimization and selection theory in mathematical finance, we define a metrics portfolio for each model as a means to aggregate a collection of metrics, and perform model selection based on the stochastic dominance of these portfolios. The statistical significance of our tests is backed theoretically by an asymptotic analysis via central limit theorems instantiated in practice via a bootstrap variance estimate. We use our framework to compare various large language models regarding risks related to drifting from instructions and outputting toxic content.
Semantic communication aims to transmit meaningful and effective information, rather than focusing on individual symbols or bits. This results in benefits like reduced latency, bandwidth usage, and higher throughput compared with traditional communication. However, semantic communication poses significant challenges due to the need for universal metrics to benchmark the joint effects of semantic information loss and practical energy consumption. This research presents a novel multi-objective loss function named "Energy-Optimized Semantic Loss" (EOSL), addressing the challenge of balancing semantic information loss and energy consumption. Through comprehensive experiments on transformer models, including CPU and GPU energy usage, it is demonstrated that EOSL-based encoder model selection can save up to 90% of energy while achieving a 44% improvement in semantic similarity performance during inference in this experiment. This work paves the way for energy-efficient neural network selection and the development of greener semantic communication architectures.
The problem of selecting a model given a set of candidates remains a challenging one that pervades many scientific fields. We employ techniques from the theory of Lie groups to analyse the symmetries in differential equation models of population growth, with the aim of informing the model selection problem. To illustrate the use of Lie symmetries in model selection, we apply them to simulated data and to coral reef data from the Great Barrier Reef, demonstrating that the trivial symmetries can distinguish between candidate models. A method for finding locally optimal parameters for multi-parameter symmetries is presented, and the paper concludes with related results, some open problems, and avenues of further research.
Keyphrase Generation (KPG) is a longstanding task in NLP with widespread applications. The advent of sequence-to-sequence (seq2seq) pre-trained language models (PLMs) has ushered in a transformative era for KPG, yielding promising performance improvements. However, many design decisions remain unexplored and are often made arbitrarily. This paper undertakes a systematic analysis of the influence of model selection and decoding strategies on PLM-based KPG. We begin by elucidating why seq2seq PLMs are apt for KPG, anchored by an attention-driven hypothesis. We then establish that conventional wisdom for selecting seq2seq PLMs lacks depth: (1) merely increasing model size or performing task-specific adaptation is not parameter-efficient; (2) although combining in-domain pre-training with task adaptation benefits KPG, it does partially hinder generalization. Regarding decoding, we demonstrate that while greedy search achieves strong F1 scores, it lags in recall compared with sampling-based methods. Based on these insights, we propose DeSel, a likelihood-based decode-select algorithm for seq2seq PLMs. DeSel improves greedy search by an average of 4.7% semantic F1 across five datasets. Our collective findings pave the way for deeper future investigations into PLM-based KPG.
Large language models (LLMs), like ChatGPT and GPT-4, have demonstrated remarkable abilities in natural language understanding and generation. However, alongside their positive impact on our daily tasks, they can also produce harmful content that negatively affects societal perceptions. To systematically assess the safety of Chinese LLMs, we introduce SuperCLUE-Safety (SC-Safety) - a multi-round adversarial benchmark with 4912 open-ended questions covering more than 20 safety sub-dimensions. Adversarial human-model interactions and conversations significantly increase the challenges compared to existing methods. Experiments on 13 major LLMs supporting Chinese yield the following insights: 1) Closed-source models outperform open-sourced ones in terms of safety; 2) Models released from China demonstrate comparable safety levels to LLMs like GPT-3.5-turbo; 3) Some smaller models with 6B-13B parameters can compete effectively in terms of safety. By introducing SC-Safety, we aim to promote collaborative efforts to create safer and more trustworthy LLMs. The benchmark and findings provide guidance on model selection. Our benchmark can be found at https://www.CLUEbenchmarks.com
Recently, Miller et al. (2021) and Baek et al. (2022) empirically demonstrated strong linear correlations between in-distribution (ID) versus out-of-distribution (OOD) accuracy and agreement. These trends, coined accuracy-on-the-line (ACL) and agreement-on-the-line (AGL), enable OOD model selection and performance estimation without labeled data. However, these phenomena also break for certain shifts, such as CIFAR10-C Gaussian Noise, posing a critical bottleneck. In this paper, we make a key finding that recent test-time adaptation (TTA) methods not only improve OOD performance, but drastically strengthen the ACL and AGL trends in models, even in shifts where models showed very weak correlations before. To analyze this, we revisit the theoretical conditions from Miller et al. (2021) that outline the types of distribution shifts needed for perfect ACL in linear models. Surprisingly, these conditions are satisfied after applying TTA to deep models in the penultimate feature embedding space. In particular, TTA causes the data distribution to collapse complex shifts into those can be expressed by a singular scaling variable in the feature space. Our results show that by combining TTA with AGL-based estimation methods, we can estimate the OOD performance of models with high precision for a broader set of distribution shifts. This lends us a simple system for selecting the best hyperparameters and adaptation strategy without any OOD labeled data.
Semantic segmentation is a critical tool in computer vision, applied in various domains like autonomous driving and medical imaging. This study focuses on aircraft contrail detection in global satellite images to improve contrail models and mitigate their impact on climate change.An innovative data preprocessing technique for NOAA GOES-16 satellite images is developed, using brightness temperature data from the infrared channel to create false-color images, enhancing model perception. To tackle class imbalance, the training dataset exclusively includes images with positive contrail labels.The model selection is based on the UPerNet architecture, implemented using the MMsegmentation library, with the integration of two ConvNeXt configurations for improved performance. Cross-entropy loss with positive class weights enhances contrail recognition. Fine-tuning employs the AdamW optimizer with a learning rate of $2.5 \times 10^{-4}$.During inference, a multi-model prediction fusion strategy and a contrail determination threshold of 0.75 yield a binary prediction mask. RLE encoding is used for efficient prediction result organization.The approach achieves exceptional results, boasting a high Dice coefficient score, placing it in the top 5\% of participating teams. This underscores the innovative nature of the segmentation model and its potential for enhanced contrail recognition in satellite imagery.For further exploration, the code and models are available on GitHub: \url{https://github.com/biluko/2023GRIC.git}.
Multilingual language models have gained significant attention in recent years, enabling the development of applications that meet diverse linguistic contexts. In this paper, we present a comprehensive evaluation of three popular multilingual language models: mBERT, XLM-R, and GPT-3. We assess their performance across a diverse set of languages, with a focus on understanding the impact of resource availability (general and model-specific), language family, script type, and word order on model performance, under two distinct tasks - text classification and text generation. Our findings reveal that while the amount of language-specific pretraining data plays a crucial role in model performance, we also identify other factors such as general resource availability, language family, and script type, as important features. We hope that our study contributes to a deeper understanding of multilingual language models to enhance their performance across languages and linguistic contexts.
Zero-shot learning in prompted vision-language models, the practice of crafting prompts to build classifiers without an explicit training process, has achieved impressive performance in many settings. This success presents a seemingly surprising observation: these methods suffer relatively little from overfitting, i.e., when a prompt is manually engineered to achieve low error on a given training set (thus rendering the method no longer actually zero-shot), the approach still performs well on held-out test data. In this paper, we show that we can explain such performance well via recourse to classical PAC-Bayes bounds. Specifically, we show that the discrete nature of prompts, combined with a PAC-Bayes prior given by a language model, results in generalization bounds that are remarkably tight by the standards of the literature: for instance, the generalization bound of an ImageNet classifier is often within a few percentage points of the true test error. We demonstrate empirically that this holds for existing handcrafted prompts and prompts generated through simple greedy search. Furthermore, the resulting bound is well-suited for model selection: the models with the best bound typically also have the best test performance. This work thus provides a possible justification for the widespread practice of prompt engineering, even if it seems that such methods could potentially overfit the training data.
Extracting medical knowledge from healthcare texts enhances downstream tasks like medical knowledge graph construction and clinical decision-making. However, the construction and application of knowledge extraction models lack automation, reusability and unified management, leading to inefficiencies for researchers and high barriers for non-AI experts such as doctors, to utilize knowledge extraction. To address these issues, we propose a ModelOps-based intelligent medical knowledge extraction framework that offers a low-code system for model selection, training, evaluation and optimization. Specifically, the framework includes a dataset abstraction mechanism based on multi-layer callback functions, a reusable model training, monitoring and management mechanism. We also propose a model recommendation method based on dataset similarity, which helps users quickly find potentially suitable models for a given dataset. Our framework provides convenience for researchers to develop models and simplifies model access for non-AI experts such as doctors.
With the emergence of large language models (LLMs), multimodal models based on LLMs have demonstrated significant potential. Models such as LLaSM, X-LLM, and SpeechGPT exhibit an impressive ability to comprehend and generate human instructions. However, their performance often falters when faced with complex tasks like end-to-end speech translation (E2E-ST), a cross-language and cross-modal translation task. In comparison to single-modal models, multimodal models lag behind in these scenarios. This paper introduces LST, a Large multimodal model designed to excel at the E2E-ST task. LST consists of a speech frontend, an adapter, and a LLM backend. The training of LST consists of two stages: (1) Modality adjustment, where the adapter is tuned to align speech representation with text embedding space, and (2) Downstream task fine-tuning, where both the adapter and LLM model are trained to optimize performance on the E2EST task. Experimental results on the MuST-C speech translation benchmark demonstrate that LST-13B achieves BLEU scores of 30.39/41.55/35.33 on En-De/En-Fr/En-Es language pairs, surpassing previous models and establishing a new state-of-the-art. Additionally, we conduct an in-depth analysis of single-modal model selection and the impact of training strategies, which lays the foundation for future research. We will open up our code and models after review.
Machine learning models have undeniably achieved impressive performance across a range of applications. However, their often perceived black-box nature, and lack of transparency in decision-making, have raised concerns about understanding their predictions. To tackle this challenge, researchers have developed methods to provide explanations for machine learning models. In this paper, we introduce LaPLACE-explainer, designed to provide probabilistic cause-and-effect explanations for any classifier operating on tabular data, in a human-understandable manner. The LaPLACE-Explainer component leverages the concept of a Markov blanket to establish statistical boundaries between relevant and non-relevant features automatically. This approach results in the automatic generation of optimal feature subsets, serving as explanations for predictions. Importantly, this eliminates the need to predetermine a fixed number N of top features as explanations, enhancing the flexibility and adaptability of our methodology. Through the incorporation of conditional probabilities, our approach offers probabilistic causal explanations and outperforms LIME and SHAP (well-known model-agnostic explainers) in terms of local accuracy and consistency of explained features. LaPLACE's soundness, consistency, local accuracy, and adaptability are rigorously validated across various classification models. Furthermore, we demonstrate the practical utility of these explanations via experiments with both simulated and real-world datasets. This encompasses addressing trust-related issues, such as evaluating prediction reliability, facilitating model selection, enhancing trustworthiness, and identifying fairness-related concerns within classifiers.
Explainable AI (XAI) methods have mostly been built to investigate and shed light on single machine learning models and are not designed to capture and explain differences between multiple models effectively. This paper addresses the challenge of understanding and explaining differences between machine learning models, which is crucial for model selection, monitoring and lifecycle management in real-world applications. We propose DeltaXplainer, a model-agnostic method for generating rule-based explanations describing the differences between two binary classifiers. To assess the effectiveness of DeltaXplainer, we conduct experiments on synthetic and real-world datasets, covering various model comparison scenarios involving different types of concept drift.
Recent advances in learning or identification of nonlinear dynamics focus on learning a suitable model within a pre-specified model class. However, a key difficulty that remains is the choice of the model class from which the dynamics will be learned. The fundamental challenge is trading the richness of the model class with the learnability within the model class. Toward addressing the so-called model selection problem, we introduce a novel notion of Structural Risk Minimization (SRM) for learning nonlinear dynamics. Inspired by classical SRM for classification, we minimize a bound on the true prediction error over hierarchies of model classes. The class selected by our SRM scheme is shown to achieve a nearly optimal learning guarantee among all model classes contained in the hierarchy. Employing the proposed scheme along with computable model class complexity bounds, we derive explicit SRM schemes for learning nonlinear dynamics under hierarchies of: i) norm-constrained Reproducing Kernel Hilbert Spaces, and ii) norm-constrained Neural Network classes. We empirically show that even though too loose to be used as absolute estimates, our SRM bounds on the true prediction error are able to track its relative behavior across different model classes of the hierarchy.
The adaptive non-harmonic (ANH) model is a powerful tool to compactly represent oscillating signals with time-varying amplitude and phase, and non-sinusoidal oscillating morphology. Given good estimators of instantaneous amplitude and phase we can construct an adaptive model, where the morphology of the oscillation is described by the wave-shape function (WSF), a 2{\pi}-periodic more general periodic function. In this paper, we address the problem of estimating the number of harmonic components of the WSF, a problem that remains underresearched, by adapting trigonometric regression model selection criteria into this context. We study the application of these criteria, originally developed in the context of stationary signals, to the case of signals with time-varying amplitudes and phases. We then incorporate the order estimation to the ANH model reconstruction procedure and analyze its performance for noisy AM-FM signals. Experimental results on synthethic signals indicate that these criteria enable the adaptive estimation of the waveform of non-stationary signals with non-sinusoidal oscillatory patterns, even in the presence of considerable amount of noise. We also apply our reconstruction procedure to the task of denoising simulated pulse wave signals and determine that the proposed technique performs competitively to other denoising schemes. We conclude this work by showing that our adaptive order estimation algorithm takes into account the interpatient waveform variability of the electrocardiogram (ECG) and respiratory signals by analyzing recordings from the Fantasia Database.
Mathematical models (MMs) are a powerful tool to help us understand and predict the dynamics of tumour growth under various conditions. In this work, we use 5 MMs with an increasing number of parameters to explore how certain (often overlooked) decisions in estimating parameters from data of experimental tumour growth affect the outcome of the analysis. In particular, we propose a framework for including tumour volume measurements that fall outside the upper and lower limits of detection, which are normally discarded. We demonstrate how excluding censored data results in an overestimation of the initial tumour volume and the MM-predicted tumour volumes prior to the first measurements, and an underestimation of the carrying capacity and the MM-predicted tumour volumes beyond the latest measurable time points. We show in which way the choice of prior for the MM parameters can impact the posterior distributions, and illustrate that reporting the highest-likelihood parameters and their 95% credible interval can lead to confusing or misleading interpretations. We hope this work will encourage others to carefully consider choices made in parameter estimation and to adopt the approaches we put forward herein.