Deep neural networks (DNNs) have long been recognized as vulnerable to backdoor attacks. By providing poisoned training data in the fine-tuning process, the attacker can implant a backdoor into the victim model. This enables input samples meeting specific textual trigger patterns to be classified as target labels of the attacker's choice. While such black-box attacks have been well explored in both computer vision and natural language processing (NLP), backdoor attacks relying on white-box attack philosophy have hardly been thoroughly investigated. In this paper, we take the first step to introduce a new type of backdoor attack that conceals itself within the underlying model architecture. Specifically, we propose to design separate backdoor modules consisting of two functions: trigger detection and noise injection. The add-on modules of model architecture layers can detect the presence of input trigger tokens and modify layer weights using Gaussian noise to disturb the feature distribution of the baseline model. We conduct extensive experiments to evaluate our attack methods using two model architecture settings on five different large language datasets. We demonstrate that the training-free architectural backdoor on a large language model poses a genuine threat. Unlike the-state-of-art work, it can survive the rigorous fine-tuning and retraining process, as well as evade output probability-based defense methods (i.e. BDDR). All the code and data is available https://github.com/SiSL-URI/Arch_Backdoor_LLM.

In this study, we leveraged Channel State Information (CSI), commonly utilized in WLAN communication, as training data to develop and evaluate five distinct machine learning models for recognizing human postures: standing, sitting, and lying down. The models we employed were: (i) Linear Discriminant Analysis, (ii) Naive Bayes-Support Vector Machine, (iii) Kernel-Support Vector Machine, (iv) Random Forest, and (v) Deep Learning. We systematically analyzed how the accuracy of these models varied with different amounts of training data. Additionally, to assess their spatial generalization capabilities, we evaluated the models' performance in a setting distinct from the one used for data collection. The experimental findings indicated that while two models -- (ii) Naive Bayes-Support Vector Machine and (v) Deep Learning -- achieved 85% or more accuracy in the original setting, their accuracy dropped to approximately 30% when applied in a different environment. These results underscore that although CSI-based machine learning models can attain high accuracy within a consistent spatial structure, their performance diminishes considerably with changes in spatial conditions, highlighting a significant challenge in their generalization capabilities.

Given $p$ samples, each of which may or may not be defective, group testing (GT) aims to determine their defect status by performing tests on $n < p$ `groups', where a group is formed by mixing a subset of the $p$ samples. Assuming that the number of defective samples is very small compared to $p$, GT algorithms have provided excellent recovery of the status of all $p$ samples with even a small number of groups. Most existing methods, however, assume that the group memberships are accurately specified. This assumption may not always be true in all applications, due to various resource constraints. Such errors could occur, eg, when a technician, preparing the groups in a laboratory, unknowingly mixes together an incorrect subset of samples as compared to what was specified. We develop a new GT method, the Debiased Robust Lasso Test Method (DRLT), that handles such group membership specification errors. The proposed DRLT method is based on an approach to debias, or reduce the inherent bias in, estimates produced by Lasso, a popular and effective sparse regression technique. We also provide theoretical upper bounds on the reconstruction error produced by our estimator. Our approach is then combined with two carefully designed hypothesis tests respectively for (i) the identification of defective samples in the presence of errors in group membership specifications, and (ii) the identification of groups with erroneous membership specifications. The DRLT approach extends the literature on bias mitigation of statistical estimators such as the LASSO, to handle the important case when some of the measurements contain outliers, due to factors such as group membership specification errors. We present numerical results which show that our approach outperforms several baselines and robust regression techniques for identification of defective samples as well as erroneously specified groups.

For electric vehicles, the Adaptive Cruise Control (ACC) in Advanced Driver Assistance Systems (ADAS) is designed to assist braking based on driving conditions, road inclines, predefined deceleration strengths, and user braking patterns. However, the driving data collected during the development of ADAS are generally limited and lack diversity. This deficiency leads to late or aggressive braking for different users. Crucially, it is necessary to effectively identify anomalies, such as unexpected or inconsistent braking patterns in ADAS, especially given the challenge of working with unlabelled, limited, and noisy datasets from real-world electric vehicles. In order to tackle the aforementioned challenges in ADAS, we propose Graph Neural Controlled Differential Equation Normalizing Flow (GDFlow), a model that leverages Normalizing Flow (NF) with Neural Controlled Differential Equations (NCDE) to learn the distribution of normal driving patterns continuously. Compared to the traditional clustering or anomaly detection algorithms, our approach effectively captures the spatio-temporal information from different sensor data and more accurately models continuous changes in driving patterns. Additionally, we introduce a quantile-based maximum likelihood objective to improve the likelihood estimate of the normal data near the boundary of the distribution, enhancing the model's ability to distinguish between normal and anomalous patterns. We validate GDFlow using real-world electric vehicle driving data that we collected from Hyundai IONIQ5 and GV80EV, achieving state-of-the-art performance compared to six baselines across four dataset configurations of different vehicle types and drivers. Furthermore, our model outperforms the latest anomaly detection methods across four time series benchmark datasets. Our approach demonstrates superior efficiency in inference time compared to existing methods.

Representing speech with discrete units has been widely used in speech codec and speech generation. However, there are several unverified claims about self-supervised discrete units, such as disentangling phonetic and speaker information with k-means, or assuming information loss after k-means. In this work, we take an information-theoretic perspective to answer how much information is present (information completeness) and how much information is accessible (information accessibility), before and after residual vector quantization. We show a lower bound for information completeness and estimate completeness on discretized HuBERT representations after residual vector quantization. We find that speaker information is sufficiently present in HuBERT discrete units, and that phonetic information is sufficiently present in the residual, showing that vector quantization does not achieve disentanglement. Our results offer a comprehensive assessment on the choice of discrete units, and suggest that a lot more information in the residual should be mined rather than discarded.

In-context learning (ICL) is a cornerstone of large language model (LLM) functionality, yet its theoretical foundations remain elusive due to the complexity of transformer architectures. In particular, most existing work only theoretically explains how the attention mechanism facilitates ICL under certain data models. It remains unclear how the other building blocks of the transformer contribute to ICL. To address this question, we study how a two-attention-layer transformer is trained to perform ICL on $n$-gram Markov chain data, where each token in the Markov chain statistically depends on the previous $n$ tokens. We analyze a sophisticated transformer model featuring relative positional embedding, multi-head softmax attention, and a feed-forward layer with normalization. We prove that the gradient flow with respect to a cross-entropy ICL loss converges to a limiting model that performs a generalized version of the induction head mechanism with a learned feature, resulting from the congruous contribution of all the building blocks. In the limiting model, the first attention layer acts as a $\mathit{copier}$, copying past tokens within a given window to each position, and the feed-forward network with normalization acts as a $\mathit{selector}$ that generates a feature vector by only looking at informationally relevant parents from the window. Finally, the second attention layer is a $\mathit{classifier}$ that compares these features with the feature at the output position, and uses the resulting similarity scores to generate the desired output. Our theory is further validated by experiments.

Recent developments in online communication and their usage in everyday life have caused an explosion in the amount of a new genre of text data, short text. Thus, the need to classify this type of text based on its content has a significant implication in many areas. Online debates are no exception, once these provide access to information about opinions, positions and preferences of its users. This paper aims to use data obtained from online social conversations in Portuguese schools (short text) to observe behavioural trends and to see if students remain engaged in the discussion when stimulated. This project used the state of the art (SoA) Machine Learning (ML) algorithms and methods, through BERT based models to classify if utterances are in or out of the debate subject. Using SBERT embeddings as a feature, with supervised learning, the proposed model achieved results above 0.95 average accuracy for classifying online messages. Such improvements can help social scientists better understand human communication, behaviour, discussion and persuasion.

Influence Functions are a standard tool for attributing predictions to training data in a principled manner and are widely used in applications such as data valuation and fairness. In this work, we present realistic incentives to manipulate influencebased attributions and investigate whether these attributions can be systematically tampered by an adversary. We show that this is indeed possible and provide efficient attacks with backward-friendly implementations. Our work raises questions on the reliability of influence-based attributions under adversarial circumstances.

It is widely acknowledged that extracting market sentiments from news data benefits market predictions. However, existing methods of using financial sentiments remain simplistic, relying on equal-weight and static aggregation to manage sentiments from multiple news items. This leads to a critical issue termed ``Aggregated Sentiment Homogenization'', which has been explored through our analysis of a large financial news dataset from industry practice. This phenomenon occurs when aggregating numerous sentiments, causing representations to converge towards the mean values of sentiment distributions and thereby smoothing out unique and important information. Consequently, the aggregated sentiment representations lose much predictive value of news data. To address this problem, we introduce the Market Attention-weighted News Aggregation Network (MANA-Net), a novel method that leverages a dynamic market-news attention mechanism to aggregate news sentiments for market prediction. MANA-Net learns the relevance of news sentiments to price changes and assigns varying weights to individual news items. By integrating the news aggregation step into the networks for market prediction, MANA-Net allows for trainable sentiment representations that are optimized directly for prediction. We evaluate MANA-Net using the S&P 500 and NASDAQ 100 indices, along with financial news spanning from 2003 to 2018. Experimental results demonstrate that MANA-Net outperforms various recent market prediction methods, enhancing Profit & Loss by 1.1% and the daily Sharpe ratio by 0.252.

One of the major challenges arising from single-cell transcriptomics experiments is the question of how to annotate the associated single-cell transcriptomic profiles. Because of the large size and the high dimensionality of the data, automated methods for annotation are needed. We focus here on datasets obtained in the context of developmental biology, where the differentiation process leads to a hierarchical structure. We consider a frequent setting where both labeled and unlabeled data are available at training time, but the sets of the labels of labeled data on one side and of the unlabeled data on the other side, are disjoint. It is an instance of the Novel Class Discovery problem. The goal is to achieve two objectives, clustering the data and mapping the clusters with labels. We propose extensions of k-Means and GMM clustering methods for solving the problem and report comparative results on artificial and experimental transcriptomic datasets. Our approaches take advantage of the hierarchical nature of the data.