Visual similarity based phishing detectors form a primary defense against credential harvesting attacks by comparing webpage appearance with legitimate references. These systems assume that brand specific visual cues such as layout structure, logo placement, and color themes are stable and difficult to replicate. We demonstrate that this assumption can be systematically violated. We present an automated, template level phishing website generation framework that produces visually realistic login pages for arbitrary brands using only the target URL. By preserving brand consistent visual cues while varying the layout and placement of credential collection elements through HTML and CSS manipulation, the framework generates phishing pages that evade visual similarity based detection. Using this approach, we curate a dataset of 1,230 phishing login pages across 123 brands and evaluate them against three state-of-the-art(SoTA) detectors: Phishlntention, PhishPedia, and an Earth Mover's Distance (EMD) based detector. The generated samples achieve up to 100% evasion against Phishlntention and the EMD based detector, and 95% evasion against PhishPedia.

Large language models (LLMs) often generate convincing, fluent explanations. However, different from humans, they often generate inconsistent explanations on different inputs. For example, an LLM may explain “all birds can fly” when answering the question “Can sparrows fly?” but meanwhile answer “no” to the related question “Can penguins fly?”. Explanations should be consistent across related examples so that they allow humans to simulate the LLM’s decision process on multiple examples. We propose explanation-consistency finetuning (EC-finetuning), a method that adapts LLMs to generate more consistent natural-language explanations on related examples. EC-finetuning involves finetuning LLMs on synthetic data that is carefully constructed to contain consistent explanations. Across a variety of question-answering datasets in various domains, EC-finetuning yields a 10.0% relative explanation consistency improvement on 4 finetuning datasets, and generalizes to 7 out-of-distribution datasets not seen during finetuning (+4.5% relative). We will make our code available for reproducibility.

The emerging capabilities of large language models (LLMs) have sparked concerns about their immediate potential for harmful misuse.The core approach to mitigate these concerns is the detection of harmful queries to the model.Current detection approaches are fallible, and are particularly susceptible to attacks that exploit mismatched generalization of model capabilities (e.g., prompts in lowresource languages or prompts provided in non-text modalities such as image and audio).To tackle this challenge, we propose OMNI-GUARD, an approach for detecting harmful prompts across languages and modalities.Our approach (i) identifies internal representations of an LLM/MLLM that are aligned across languages or modalities and then (ii) uses them to build a language-agnostic or modality-agnostic classifier for detecting harmful prompts.OM-NIGUARD improves harmful prompt classification accuracy by 11.57% over the strongest baseline in a multilingual setting, by 20.44% for image-based prompts, and sets a new SOTA for audio-based prompts.By repurposing embeddings computed during generation, OMNI-GUARD is also very efficient (≈ 120× faster than the next fastest baseline).Code and data are available at https://github.com/ vsahil/OmniGuard.

Large language models (LLMs) have shown remarkable in-context learning (ICL) capabilities on textual data. We explore whether these capabilities can be extended to continuous vectors from diverse domains, obtained from black-box pretrained encoders. By aligning input data with an LLM's embedding space through lightweight projectors, we observe that LLMs can effectively process and learn from these projected vectors, which we term Vector-ICL. In particular, we find that pretraining projectors with general language modeling objectives enables Vector-ICL, while task-specific finetuning further enhances performance. In our experiments across various tasks and modalities, including text reconstruction, numerical function regression, text classification, summarization, molecule captioning, time-series classification, graph classification, and fMRI decoding, Vector-ICL often surpasses both few-shot ICL and domain-specific model or tuning. We further conduct analyses and case studies, indicating the potential of LLMs to process vector representations beyond traditional token-based paradigms.

DINO and DINOv2 are two model families being widely used to learn representations from unlabeled imagery data at large scales. Their learned representations often enable state-of-the-art performance for downstream tasks, such as image classification and segmentation. However, they employ many empirically motivated design choices and their training pipelines are highly complex and unstable — many hyperparameters need to be carefully tuned to ensure that the representations do not collapse — which poses considerable difficulty to improving them or adapting them to new domains. In this work, we posit that we can remove most such-motivated idiosyncrasies in the pre-training pipelines, and only need to add an explicit coding rate term in the loss function to avoid collapse of the representations. As a result, we obtain highly simplified variants of the DINO and DINOv2 which we call SimDINO and SimDINOv2, respectively. Remarkably, these simplified models are more robust to different design choices, such as network architecture and hyperparameters, and they learn even higher-quality representations, measured by performance on downstream tasks, offering a Pareto improvement over the corresponding DINO and DINOv2 models. This work highlights the potential of using simplifying design principles to improve the empirical practice of deep learning. Code and model checkpoints are available at https://github.com/RobinWu218/SimDINO.

Concept Bottleneck Models (CBMs) have been proposed as a compromise between white-box and black-box models, aiming to achieve interpretability without sacrificing accuracy. The standard training procedure for CBMs is to predefine a candidate set of human-interpretable concepts, extract their values from the training data, and identify a sparse subset as inputs to a transparent prediction model. However, such approaches are often hampered by the tradeoff between exploring a sufficiently large set of concepts versus controlling the cost of obtaining concept extractions, resulting in a large interpretability-accuracy tradeoff. This work investigates a novel approach that sidesteps these challenges: BC-LLM iteratively searches over a potentially infinite set of concepts within a Bayesian framework, in which Large Language Models (LLMs) serve as both a concept extraction mechanism and prior. Even though LLMs can be miscalibrated and hallucinate, we prove that BC-LLM can provide rigorous statistical inference and uncertainty quantification. Across image, text, and tabular datasets, BC-LLM outperforms interpretable baselines and even black-box models in certain settings, converges more rapidly towards relevant concepts, and is more robust to out-of-distribution samples.

While large transformer models excel in predictive performance, their lack of interpretability restricts their usefulness in high-stakes domains. To remedy this, we propose the Generalized Induction-Head Model (GIM), an interpretable model for next-token prediction inspired by the observation of “induction heads” in LLMs. GIM is a retrieval-based module that identifies similar sequences in the input context by combining exact n-gram matching and fuzzy matching based on a neural similarity metric. We evaluate GIM in two settings: language modeling and fMRI response prediction. In language modeling, GIM improves next-token prediction by up to 25%p over interpretable baselines, significantly narrowing the gap with black-box LLMs. In an fMRI setting, GIM improves neural response prediction by 20% and offers insights into the language selectivity of the brain. GIM represents a significant step toward uniting interpretability and performance across domains. The code is available at https://github.com/ejkim47/generalized-induction-head.

In standard autoregressive generation, an LLM predicts the next-token distribution, samples a discrete token, and then discards the distribution, passing only the sampled token as new input. To preserve this distribution’s rich information, we propose Mixture of Inputs (MoI), a training-free method for autoregressive generation. After generating a token following the standard paradigm, we construct a new input that blends the generated discrete token with the previously discarded token distribution. Specifically, we employ a Bayesian estimation method that treats the token distribution as the prior, the sampled token as the observation, and replaces the conventional one-hot vector with the continuous posterior expectation as the new model input. MoI allows the model to maintain a richer internal representation throughout the generation process, resulting in improved text quality and reasoning capabilities. On mathematical reasoning, code generation, and PhD-level QA tasks, MoI consistently improves performance across multiple models including QwQ-32B, Nemotron-Super-49B, Gemma-3-27B, and DAPO-Qwen-32B, with no additional training and negligible computational overhead.

In human-written articles, we often leverage the subtleties of text style, such as bold and italics, to guide the attention of readers. These textual emphases are vital for the readers to grasp the conveyed information. When interacting with large language models (LLMs), we have a similar need -- steering the model to pay closer attention to user-specified information, e.g., an instruction. Existing methods, however, are constrained to process plain text and do not support such a mechanism. This motivates us to introduce PASTA -- Post-hoc Attention STeering Approach, a method that allows LLMs to read text with user-specified emphasis marks. To this end, PASTA identifies a small subset of attention heads and applies precise attention reweighting on them, directing the model attention to user-specified parts. Like prompting, PASTA is applied at inference time and does not require changing any model parameters. Experiments demonstrate that PASTA can substantially enhance an LLM's ability to follow user instructions or integrate new knowledge from user inputs, leading to a significant performance improvement on a variety of tasks, e.g., an average accuracy improvement of 22\% for LLAMA-7B. Our code is publicly available at https://github.com/QingruZhang/PASTA .

Large language models (LLMs) have rapidly improved text embeddings for a growing array of natural-language processing tasks. However, their opaqueness and proliferation into scientific domains such as neuroscience have created a growing need for interpretability. Here, we ask whether we can obtain interpretable embeddings through LLM prompting. We introduce question-answering embeddings (QA-Emb), embeddings where each feature represents an answer to a yes/no question asked to an LLM. Training QA-Emb reduces to selecting a set of underlying questions rather than learning model weights. We use QA-Emb to flexibly generate interpretable models for predicting fMRI voxel responses to language stimuli. QA-Emb significantly outperforms an established interpretable baseline, and does so while requiring very few questions. This paves the way towards building flexible feature spaces that can concretize and evaluate our understanding of semantic brain representations. We additionally find that QA-Emb can be effectively approximated with an efficient model, and we explore broader applications in simple NLP tasks.

Prompting language models (LMs) is the main interface for applying them to new tasks.However, for smaller LMs, prompting provides low accuracy compared to gradient-based finetuning.Tree Prompting is an approach to prompting which builds a decision tree of prompts, linking multiple LM calls together to solve a task.At inference time, each call to the LM is determined by efficiently routing the outcome of the previous call using the tree.Experiments on classification datasets show that Tree Prompting improves accuracy over competing methods and is competitive with fine-tuning.We also show that variants of Tree Prompting allow inspection of a model's decision-making process. 1

The convolutional layers of the standard convolutional neural networks (CNNs) are equivariant to translation. Recently, a new class of CNNs is introduced which is equivariant to other affine geometric transformations such as rotation and reflection by replacing the standard convolutional layer with the group convolutional layer or using the steerable filters in the convloutional layer. We propose to embed the 2D positional encoding which is invariant to rotation, reflection and translation using orthogonal polar harmonic transforms (PHTs) before flattening the feature maps for fully-connected or classification layer in the equivariant CNN architecture. We select the PHTs among several invariant transforms, as they are very efficient in performance and speed. The proposed 2D positional encoding scheme between the convolutional and fully-connected layers of the equivariant networks is shown to provide significant improvement in performance on the rotated MNIST, CIFAR-10 and CIFAR-100 datasets.

Complexity is a fundamental concept underlying statistical learning theory that aims to inform generalization performance. Parameter count, while successful in low-dimensional settings, is not well-justified for overparameterized settings when the number of parameters is more than the number of training samples. We revisit complexity measures based on Rissanen's principle of minimum description length (MDL) and define a novel MDL-based complexity (MDL-COMP) that remains valid for overparameterized models. MDL-COMP is defined via an optimality criterion over the encodings induced by a good Ridge estimator class. We provide an extensive theoretical characterization of MDL-COMP for linear models and kernel methods and show that it is not just a function of parameter count, but rather a function of the singular values of the design or the kernel matrix and the signal-to-noise ratio. For a linear model with $n$ observations, $d$ parameters, and i.i.d. Gaussian predictors, MDL-COMP scales linearly with $d$ when $dn$. For kernel methods, we show that MDL-COMP informs minimax in-sample error, and can decrease as the dimensionality of the input increases. We also prove that MDL-COMP upper bounds the in-sample mean squared error (MSE). Via an array of simulations and real-data experiments, we show that a data-driven Prac-MDL-COMP informs hyper-parameter tuning for optimizing test MSE with ridge regression in limited data settings, sometimes improving upon cross-validation and (always) saving computational costs. Finally, our findings also suggest that the recently observed double decent phenomenons in overparameterized models might be a consequence of the choice of non-ideal estimators. [abs][pdf][bib] [code] © JMLR 2023. (edit, beta) Mastodon

Decision trees and random forests (RF) are a cornerstone of modern machine learning practice. Due to their tendency to overfit, trees are typically regularized by a variety of techniques that modify their structure (e.g. pruning). We introduce Hierarchical Shrinkage (HS), a post-hoc algorithm which regularizes the tree not by altering its structure, but by shrinking the prediction over each leaf toward the sample means over each of its ancestors, with weights depending on a single regularization parameter and the number of samples in each ancestor. Since HS is a post-hoc method, it is extremely fast, compatible with any tree-growing algorithm and can be used synergistically with other regularization techniques. Extensive experiments over a wide variety of real-world datasets show that HS substantially increases the predictive performance of decision trees even when used in conjunction with other regularization techniques. Moreover, we find that applying HS to individual trees in a RF often improves its accuracy and interpretability by simplifying and stabilizing decision boundaries and SHAP values. We further explain HS by showing that it to be equivalent to ridge regression on a basis that is constructed of decision stumps associated to the internal nodes of a tree. All code and models are released in a full-fledged package available on Github

The convolutional layers of standard convolutional neural networks (CNNs) are equivariant to translation. However, the convolution and fully-connected layers are not equivariant or invariant to other affine geometric transformations. Recently, a new class of CNNs is proposed in which the conventional layers of CNNs are replaced with equivariant convolution, pooling, and batch-normalization layers. The final classification layer in equivariant neural networks is invariant to different affine geometric transformations such as rotation, reflection and translation, and the scalar value is obtained by either eliminating the spatial dimensions of filter responses using convolution and down-sampling throughout the network or average is taken over the filter responses. In this work, we propose to integrate the orthogonal moments which gives the high-order statistics of the function as an effective means for encoding global invariance with respect to rotation, reflection and translation in fully-connected layers. As a result, the intermediate layers of the network become equivariant while the classification layer becomes invariant. The most widely used Zernike, pseudo-Zernike and orthogonal Fourier-Mellin moments are considered for this purpose. The effectiveness of the proposed work is evaluated by integrating the invariant transition and fully-connected layer in the architecture of group-equivariant CNNs (G-CNNs) on rotated MNIST and CIFAR10 datasets.

Recent deep-learning models have achieved impressive prediction performance, but often sacrifice interpretability and computational efficiency. Interpretability is crucial in many disciplines, such as science and medicine, where models must be carefully vetted or where interpretation is the goal itself. Moreover, interpretable models are concise and often yield computational efficiency. Here, we propose adaptive wavelet distillation (AWD), a method which aims to distill information from a trained neural network into a wavelet transform. Specifically, AWD penalizes feature attributions of a neural network in the wavelet domain to learn an effective multi-resolution wavelet transform. The resulting model is highly predictive, concise, computationally efficient, and has properties (such as a multi-scale structure) which make it easy to interpret. In close collaboration with domain experts, we showcase how AWD addresses challenges in two real-world settings: cosmological parameter inference and molecular-partner prediction. In both cases, AWD yields a scientifically interpretable and concise model which gives predictive performance better than state-of-the-art neural networks. Moreover, AWD identifies predictive features that are scientifically meaningful in the context of respective domains. All code and models are released in a full-fledged package available on Github.

For an explanation of a deep learning model to be effective, it must provide both insight into a model and suggest a corresponding action in order to achieve some objective. Too often, the litany of proposed explainable deep learning methods stop at the first step, providing practitioners with insight into a model, but no way to act on it. In this paper, we propose contextual decomposition explanation penalization (CDEP), a method which enables practitioners to leverage existing explanation methods to increase the predictive accuracy of a deep learning model. In particular, when shown that a model has incorrectly assigned importance to some features, CDEP enables practitioners to correct these errors by inserting domain knowledge into the model via explanations. We demonstrate the ability of CDEP to increase performance on an array of toy and real datasets.

We present a method combining affinity prediction with region agglomeration, which improves significantly upon the state of the art of neuron segmentation from electron microscopy (EM) in accuracy and scalability. Our method consists of a 3D U-Net, trained to predict affinities between voxels, followed by iterative region agglomeration. We train using a structured loss based on Malis, encouraging topologically correct segmentations obtained from affinity thresholding. Our extension consists of two parts: First, we present a quasi-linear method to compute the loss gradient, improving over the original quadratic algorithm. Second, we compute the gradient in two separate passes to avoid spurious gradient contributions in early training stages. Our predictions are accurate enough that simple learning-free percentile-based agglomeration outperforms more involved methods used earlier on inferior predictions. We present results on three diverse EM datasets, achieving relative improvements over previous results of 27, 15, and 250 percent. Our findings suggest that a single method can be applied to both nearly isotropic block-face EM data and anisotropic serial sectioned EM data. The runtime of our method scales linearly with the size of the volume and achieves a throughput of $\sim$∼ 2.6 seconds per megavoxel, qualifying our method for the processing of very large datasets.

The authors propose a novel image adaptive watermarking scheme for geometrically invariant and high‐capacity data embedding scheme based on accurate and fast framework for the computation of Zernike moments (ZMs). The high capacity is achieved by maximising data embedding size and improving the hiding ratio after reducing the inaccuracies in the computation of ZMs. Furthermore, they also introduce a concept of conditional quantisation technique which enables to reduce the total number of ZMs needed to be modified during watermark embedding. This concept enhances the visual imperceptibility of the watermarked image and its robustness against various attacks. Numerous experiments have been performed to demonstrate the improvements in embedding capacity, visual imperceptibility, watermark robustness and time complexity as compared with the existing methods.

A polar system is used to compute Zernike moments to enhance their accuracy and to improve invariance to rotation. This requires reconfiguration of pixel arrangements that are normally available in rectangular grids in the cartesian coordinate system. This study presents an improved reconfiguration model of pixel arrangements that uses nearly 27.3% less number of pixels compared to the existing model, thus enhancing the computational efficiency of the proposed method by the same percentage. The performance of the proposed model is analysed in detail, which is observed to be at par with the existing method.

A shape based post processing system for an OCR of Gurmukhi script has been developed. Based on the size and shape of a word, the Punjabi corpora has been split into different partitions. The statistical information of Punjabi language syllable combination, corpora look up and holistic recognition of most commonly occurring words have been combined to design the post processor. An improvement of 3% in recognition rate from 94.35% to 97.34% has been reported on machine printed images using the post processing techniques.

A system for recognition of machine printed Gurmukhi script is presented. The recognition system presented operates at sub-character level. The segmentation process breaks a word into sub-characters and the recognition phase consists of classifying these sub-characters and combining them to form Gurmukhi characters. A set of very simple and easy to computer features is used and a hybrid classification scheme consisting of binary decision trees and nearest neighbours is employed. A recognition rate of 96.6% at the processing speed of 175 characters second was achieved on clean images of text without employing any post-processing technique.