Research connected to &quot;machine learning&quot;

Sequence Parallelism: Long Sequence Training from System Perspective

Transformer achieves promising results on various tasks. However, self-attention suffers from quadratic memory requirements with respect to the sequence length. Existing work focuses on reducing time and space complexity from an algorithm perspective. In this work, we propose sequence parallelism, a memory-efficient parallelism method to help us break input sequence length limitation and train with longer sequences on GPUs efficiently. Our approach is compatible with most existing parallelisms (e.g. data parallelism, pipeline parallelism and tensor parallelism), which means our sequence parallelism makes 4D parallelism possible. More importantly, we no longer require a single device to hold the whole sequence. That is, with sparse attention, our sequence parallelism enables us to train transformer with infinite long sequence. Specifically, we split the input sequence into multiple chunks and feed each chunk into its corresponding device (i.e. GPU). To compute the attention output, we integrated ring-style communication with self-attention calculation and proposed Ring Self-Attention (RSA). Experiments show that sequence parallelism performs well when scaling with batch size and sequence length. Compared with tensor parallelism, our approach achieved $13.7\times$ and $3.0\times$ maximum batch size and sequence length respectively when scaling up to 64 NVIDIA P100 GPUs. With sparse attention, sequence can handle sequence with over 114K tokens, which is over $27\times$ longer than existing sparse attention works holding the whole sequence on a single device.

Anomaly Detection for Unmanned Aerial Vehicle Sensor Data Using a Stacked Recurrent Autoencoder Method with Dynamic Thresholding

With substantial recent developments in aviation technologies, Unmanned Aerial Vehicles (UAVs) are becoming increasingly integrated in commercial and military operations internationally. Research into the applications of aircraft data is essential in improving safety, reducing operational costs, and developing the next frontier of aerial technology. Having an outlier detection system that can accurately identify anomalous behaviour in aircraft is crucial for these reasons. This paper proposes a system incorporating a Long Short-Term Memory (LSTM) Deep Learning Autoencoder based method with a novel dynamic thresholding algorithm and weighted loss function for anomaly detection of a UAV dataset, in order to contribute to the ongoing efforts that leverage innovations in machine learning and data analysis within the aviation industry. The dynamic thresholding and weighted loss functions showed promising improvements to the standard static thresholding method, both in accuracy-related performance metrics and in speed of true fault detection.

Reward Machines: Exploiting Reward Function Structure in Reinforcement Learning

Reinforcement learning (RL) methods usually treat reward functions as black boxes. As such, these methods must extensively interact with the environment in order to discover rewards and optimal policies. In most RL applications, however, users have to program the reward function and, hence, there is the opportunity to make the reward function visible -- to show the reward function's code to the RL agent so it can exploit the function's internal structure to learn optimal policies in a more sample efficient manner. In this paper, we show how to accomplish this idea in two steps. First, we propose reward machines, a type of finite state machine that supports the specification of reward functions while exposing reward function structure. We then describe different methodologies to exploit this structure to support learning, including automated reward shaping, task decomposition, and counterfactual reasoning with off-policy learning. Experiments on tabular and continuous domains, across different tasks and RL agents, show the benefits of exploiting reward structure with respect to sample efficiency and the quality of resultant policies. Finally, by virtue of being a form of finite state machine, reward machines have the expressive power of a regular language and as such support loops, sequences and conditionals, as well as the expression of temporally extended properties typical of linear temporal logic and non-Markovian reward specification.

preprint2014arXiv

Grid Topology Identification using Electricity Prices

The potential of recovering the topology of a grid using solely publicly available market data is explored here. In contemporary whole-sale electricity markets, real-time prices are typically determined by solving the network-constrained economic dispatch problem. Under a linear DC model, locational marginal prices (LMPs) correspond to the Lagrange multipliers of the linear program involved. The interesting observation here is that the matrix of spatiotemporally varying LMPs exhibits the following property: Once premultiplied by the weighted grid Laplacian, it yields a low-rank and sparse matrix. Leveraging this rich structure, a regularized maximum likelihood estimator (MLE) is developed to recover the grid Laplacian from the LMPs. The convex optimization problem formulated includes low rank- and sparsity-promoting regularizers, and it is solved using a scalable algorithm. Numerical tests on prices generated for the IEEE 14-bus benchmark provide encouraging topology recovery results.

preprint2016arXiv

Masking Strategies for Image Manifolds

We consider the problem of selecting an optimal mask for an image manifold, i.e., choosing a subset of the pixels of the image that preserves the manifold's geometric structure present in the original data. Such masking implements a form of compressive sensing through emerging imaging sensor platforms for which the power expense grows with the number of pixels acquired. Our goal is for the manifold learned from masked images to resemble its full image counterpart as closely as possible. More precisely, we show that one can indeed accurately learn an image manifold without having to consider a large majority of the image pixels. In doing so, we consider two masking methods that preserve the local and global geometric structure of the manifold, respectively. In each case, the process of finding the optimal masking pattern can be cast as a binary integer program, which is computationally expensive but can be approximated by a fast greedy algorithm. Numerical experiments show that the relevant manifold structure is preserved through the data-dependent masking process, even for modest mask sizes.

preprint2014arXiv

Short-Term Predictability of Photovoltaic Production over Italy

Photovoltaic (PV) power production increased drastically in Europe throughout the last years. About the 6% of electricity in Italy comes from PV and for an efficient management of the power grid an accurate and reliable forecasting of production would be needed. Starting from a dataset of electricity production of 65 Italian solar plants for the years 2011-2012 we investigate the possibility to forecast daily production from one to ten days of lead time without using on site measurements. Our study is divided in two parts: an assessment of the predictability of meteorological variables using weather forecasts and an analysis on the application of data-driven modelling in predicting solar power production. We calibrate a SVM model using available observations and then we force the same model with the predicted variables from weather forecasts with a lead time from one to ten days. As expected, solar power production is strongly influenced by cloudiness and clear sky, in fact we observe that while during summer we obtain a general error under the 10% (slightly lower in south Italy), during winter the error is abundantly above the 20%.

Scaling Laws for Mixture Pretraining Under Data Constraints

As language models scale, the amount of data they require grows -- yet many target data sources, such as low-resource languages or specialized domains, are inherently limited in size. A common strategy is to mix this scarce but valuable target data with abundant generic data, which presents a fundamental trade-off: too little target data in the mixture underexposes the model to the target domain, while too much target data repeats the same examples excessively, yielding diminishing returns and eventual overfitting. We study this trade-off across more than 2,000 language-model training runs spanning multiple model and target dataset sizes, as well as several data types, including multilingual, domain-specific, and quality-filtered mixtures. Across all settings, we find that repetition is a central driver of target-domain performance, and that mixture training tolerates much higher repetition than single-source training: scarce target corpora can be reused 15-20 times, with the optimal number of repetitions depending on the target data size, compute budget, and model scale. Next, we introduce a repetition-aware mixture scaling law that accounts for the decreasing value of repeated target tokens and the regularizing role of generic data. Optimizing the scaling law provides a principled way to compute effective mixture configurations, yielding practical mixture recommendations for pretraining under data constraints.

Few-Shot Forecasting of Time-Series with Heterogeneous Channels

Learning complex time series forecasting models usually requires a large amount of data, as each model is trained from scratch for each task/data set. Leveraging learning experience with similar datasets is a well-established technique for classification problems called few-shot classification. However, existing approaches cannot be applied to time-series forecasting because i) multivariate time-series datasets have different channels and ii) forecasting is principally different from classification. In this paper we formalize the problem of few-shot forecasting of time-series with heterogeneous channels for the first time. Extending recent work on heterogeneous attributes in vector data, we develop a model composed of permutation-invariant deep set-blocks which incorporate a temporal embedding. We assemble the first meta-dataset of 40 multivariate time-series datasets and show through experiments that our model provides a good generalization, outperforming baselines carried over from simpler scenarios that either fail to learn across tasks or miss temporal information.

preprint2021arXiv

Selecting the suitable resampling strategy for imbalanced data classification regarding dataset properties

In many application domains such as medicine, information retrieval, cybersecurity, social media, etc., datasets used for inducing classification models often have an unequal distribution of the instances of each class. This situation, known as imbalanced data classification, causes low predictive performance for the minority class examples. Thus, the prediction model is unreliable although the overall model accuracy can be acceptable. Oversampling and undersampling techniques are well-known strategies to deal with this problem by balancing the number of examples of each class. However, their effectiveness depends on several factors mainly related to data intrinsic characteristics, such as imbalance ratio, dataset size and dimensionality, overlapping between classes or borderline examples. In this work, the impact of these factors is analyzed through a comprehensive comparative study involving 40 datasets from different application areas. The objective is to obtain models for automatic selection of the best resampling strategy for any dataset based on its characteristics. These models allow us to check several factors simultaneously considering a wide range of values since they are induced from very varied datasets that cover a broad spectrum of conditions. This differs from most studies that focus on the individual analysis of the characteristics or cover a small range of values. In addition, the study encompasses both basic and advanced resampling strategies that are evaluated by means of eight different performance metrics, including new measures specifically designed for imbalanced data classification. The general nature of the proposal allows the choice of the most appropriate method regardless of the domain, avoiding the search for special purpose techniques that could be valid for the target data.

BROS: Bias-Corrected Randomized Subspaces for Memory-Efficient Single-Loop Bilevel Optimization

Stochastic bilevel optimization (SBO) has become a standard framework for hyperparameter learning, data reweighting, representation learning, and data-mixture optimization in deep learning. Existing exact single-loop SBO methods and memory-efficient surrogate SBO methods either create severe memory pressure for large lower-level neural networks or lack competitive convergence guarantees under standard assumptions. In this paper, we propose BROS, a memory-efficient single-loop SBO method with the same convergence rate order as exact single-loop SBO methods. BROS performs lower and auxiliary updates in randomized subspaces with a Rademacher bi-probe correction that recovers an unbiased Hessian-action estimator. We prove that BROS preserves the $\mathcal O(\varepsilon^{-2})$ sample complexity of MA-SOBA for finding an $\varepsilon$-stationary point under only standard assumptions. Experiments on hyper-data cleaning, data-mixture learning, hyper-representation learning, and ViT sample reweighting show that BROS reduces peak memory by up to 44.9% while closely matching full-space baseline performance.

LTU Attacker for Membership Inference

We address the problem of defending predictive models, such as machine learning classifiers (Defender models), against membership inference attacks, in both the black-box and white-box setting, when the trainer and the trained model are publicly released. The Defender aims at optimizing a dual objective: utility and privacy. Both utility and privacy are evaluated with an external apparatus including an Attacker and an Evaluator. On one hand, Reserved data, distributed similarly to the Defender training data, is used to evaluate Utility; on the other hand, Reserved data, mixed with Defender training data, is used to evaluate membership inference attack robustness. In both cases classification accuracy or error rate are used as the metric: Utility is evaluated with the classification accuracy of the Defender model; Privacy is evaluated with the membership prediction error of a so-called "Leave-Two-Unlabeled" LTU Attacker, having access to all of the Defender and Reserved data, except for the membership label of one sample from each. We prove that, under certain conditions, even a "naïve" LTU Attacker can achieve lower bounds on privacy loss with simple attack strategies, leading to concrete necessary conditions to protect privacy, including: preventing over-fitting and adding some amount of randomness. However, we also show that such a naïve LTU Attacker can fail to attack the privacy of models known to be vulnerable in the literature, demonstrating that knowledge must be complemented with strong attack strategies to turn the LTU Attacker into a powerful means of evaluating privacy. Our experiments on the QMNIST and CIFAR-10 datasets validate our theoretical results and confirm the roles of over-fitting prevention and randomness in the algorithms to protect against privacy attacks.

Dywave: Event-Aligned Dynamic Tokenization for Heterogeneous IoT Sensing Signals

Internet of Things (IoT) systems continuously collect heterogeneous sensing signals from ubiquitous sensors to support intelligent applications such as human activity analysis, emotion monitoring, and environmental perception. These signals are inherently non-stationary and multi-scale, posing unique challenges for standard tokenization techniques. This paper proposes Dywave, a dynamic tokenization framework for IoT sensing signals that constructs compact input representations aligned with intrinsic temporal structures and underlying physical events. Dywave leverages wavelet-based hierarchical decomposition, identifies meaningful temporal boundaries corresponding to underlying semantic events, and adaptively compresses redundant intervals while preserving temporal coherence. Extensive evaluations on five real-world IoT sensing datasets across activity recognition, stress assessment, and nearby object detection demonstrate that Dywave outperforms state-of-the-art methods by up to 12% in accuracy, while improving computational efficiency by reducing input token lengths by up to 75% across mainstream sequence models. Moreover, Dywave exhibits improved robustness to domain shifts and varying sequence lengths.

Comparative Analysis of the Hidden Markov Model and LSTM: A Simulative Approach

Time series and sequential data have gained significant attention recently since many real-world processes in various domains such as finance, education, biology, and engineering can be modeled as time series. Although many algorithms and methods such as the Kalman filter, hidden Markov model, and long short term memory (LSTM) are proposed to make inferences and predictions for the data, their usage significantly depends on the application, type of the problem, available data, and sufficient accuracy or loss. In this paper, we compare the supervised and unsupervised hidden Markov model to LSTM in terms of the amount of data needed for training, complexity, and forecasting accuracy. Moreover, we propose various techniques to discretize the observations and convert the problem to a discrete hidden Markov model under stationary and non-stationary situations. Our results indicate that even an unsupervised hidden Markov model can outperform LSTM when a massive amount of labeled data is not available. Furthermore, we show that the hidden Markov model can still be an effective method to process the sequence data even when the first-order Markov assumption is not satisfied.

Uncertainty in Contrastive Learning: On the Predictability of Downstream Performance

The superior performance of some of today's state-of-the-art deep learning models is to some extent owed to extensive (self-)supervised contrastive pretraining on large-scale datasets. In contrastive learning, the network is presented with pairs of positive (similar) and negative (dissimilar) datapoints and is trained to find an embedding vector for each datapoint, i.e., a representation, which can be further fine-tuned for various downstream tasks. In order to safely deploy these models in critical decision-making systems, it is crucial to equip them with a measure of their uncertainty or reliability. However, due to the pairwise nature of training a contrastive model, and the lack of absolute labels on the output (an abstract embedding vector), adapting conventional uncertainty estimation techniques to such models is non-trivial. In this work, we study whether the uncertainty of such a representation can be quantified for a single datapoint in a meaningful way. In other words, we explore if the downstream performance on a given datapoint is predictable, directly from its pre-trained embedding. We show that this goal can be achieved by directly estimating the distribution of the training data in the embedding space and accounting for the local consistency of the representations. Our experiments show that this notion of uncertainty for an embedding vector often strongly correlates with its downstream accuracy.

Divergence-aware Federated Self-Supervised Learning

Self-supervised learning (SSL) is capable of learning remarkable representations from centrally available data. Recent works further implement federated learning with SSL to learn from rapidly growing decentralized unlabeled images (e.g., from cameras and phones), often resulted from privacy constraints. Extensive attention has been paid to SSL approaches based on Siamese networks. However, such an effort has not yet revealed deep insights into various fundamental building blocks for the federated self-supervised learning (FedSSL) architecture. We aim to fill in this gap via in-depth empirical study and propose a new method to tackle the non-independently and identically distributed (non-IID) data problem of decentralized data. Firstly, we introduce a generalized FedSSL framework that embraces existing SSL methods based on Siamese networks and presents flexibility catering to future methods. In this framework, a server coordinates multiple clients to conduct SSL training and periodically updates local models of clients with the aggregated global model. Using the framework, our study uncovers unique insights of FedSSL: 1) stop-gradient operation, previously reported to be essential, is not always necessary in FedSSL; 2) retaining local knowledge of clients in FedSSL is particularly beneficial for non-IID data. Inspired by the insights, we then propose a new approach for model update, Federated Divergence-aware Exponential Moving Average update (FedEMA). FedEMA updates local models of clients adaptively using EMA of the global model, where the decay rate is dynamically measured by model divergence. Extensive experiments demonstrate that FedEMA outperforms existing methods by 3-4% on linear evaluation. We hope that this work will provide useful insights for future research.

Beyond Oversquashing: Understanding Signal Propagation in GNNs Via Observables

Graph Neural Networks (GNNs) perform computations on graphs by routing the signal between graph regions using a graph shift operator or a message passing scheme. Often, the propagation of the signal leads to a loss of information, where the signal tends to diffuse across the graph instead of being deliberately routed between regions of interest. Two notions that depict this phenomenon are oversmoothing and oversquashing. In this paper, we propose an alternative approach for modeling signal propagation, inspired by quantum mechanics, using the notion of observables. Specifically, we model the place in the graph where the signal lies, how much the signal is concentrated there, and how much of the signal is propagated towards a location of interest when applying a GNN. Using these new concepts, we prove that standard spectral GNNs have poor signal propagation capabilities. We then propose a new type of spectral GNN, termed Schrödinger GNN, which we show has a superior capacity to route the signal across the graph.

Generating Higher-Fidelity Synthetic Datasets with Privacy Guarantees

This paper considers the problem of enhancing user privacy in common machine learning development tasks, such as data annotation and inspection, by substituting the real data with samples form a generative adversarial network. We propose employing Bayesian differential privacy as the means to achieve a rigorous theoretical guarantee while providing a better privacy-utility trade-off. We demonstrate experimentally that our approach produces higher-fidelity samples, compared to prior work, allowing to (1) detect more subtle data errors and biases, and (2) reduce the need for real data labelling by achieving high accuracy when training directly on artificial samples.

preprint2014arXiv

Active Learning for Undirected Graphical Model Selection

This paper studies graphical model selection, i.e., the problem of estimating a graph of statistical relationships among a collection of random variables. Conventional graphical model selection algorithms are passive, i.e., they require all the measurements to have been collected before processing begins. We propose an active learning algorithm that uses junction tree representations to adapt future measurements based on the information gathered from prior measurements. We prove that, under certain conditions, our active learning algorithm requires fewer scalar measurements than any passive algorithm to reliably estimate a graph. A range of numerical results validate our theory and demonstrates the benefits of active learning.

Geometric Insights into the Convergence of Nonlinear TD Learning

While there are convergence guarantees for temporal difference (TD) learning when using linear function approximators, the situation for nonlinear models is far less understood, and divergent examples are known. Here we take a first step towards extending theoretical convergence guarantees to TD learning with nonlinear function approximation. More precisely, we consider the expected learning dynamics of the TD(0) algorithm for value estimation. As the step-size converges to zero, these dynamics are defined by a nonlinear ODE which depends on the geometry of the space of function approximators, the structure of the underlying Markov chain, and their interaction. We find a set of function approximators that includes ReLU networks and has geometry amenable to TD learning regardless of environment, so that the solution performs about as well as linear TD in the worst case. Then, we show how environments that are more reversible induce dynamics that are better for TD learning and prove global convergence to the true value function for well-conditioned function approximators. Finally, we generalize a divergent counterexample to a family of divergent problems to demonstrate how the intera

Interestingness Elements for Explainable Reinforcement Learning: Understanding Agents' Capabilities and Limitations

We propose an explainable reinforcement learning (XRL) framework that analyzes an agent's history of interaction with the environment to extract interestingness elements that help explain its behavior. The framework relies on data readily available from standard RL algorithms, augmented with data that can easily be collected by the agent while learning. We describe how to create visual summaries of an agent's behavior in the form of short video-clips highlighting key interaction moments, based on the proposed elements. We also report on a user study where we evaluated the ability of humans to correctly perceive the aptitude of agents with different characteristics, including their capabilities and limitations, given visual summaries automatically generated by our framework. The results show that the diversity of aspects captured by the different interestingness elements is crucial to help humans correctly understand an agent's strengths and limitations in performing a task, and determine when it might need adjustments to improve its performance.

Spectral Gradient Surgery for Domain-Generalizable Dataset Distillation

Dataset Distillation (DD) synthesizes a compact synthetic dataset that preserves the training utility of a full dataset. However, its standard formulation assumes that test data follow the same distribution as training data, an assumption that rarely holds in practice. A straightforward extension-applying post-hoc Domain Generalization (DG) techniques to distilled data-is ill-suited because existing DG methods rely on the natural diversity of real datasets, which compact synthetic sets inherently lack, while also incurring substantial augmentation overhead that conflicts with the efficiency objective of dataset distillation. To address this limitation, we introduce Domain Generalizable Dataset Distillation (DGDD), a new problem setting that explicitly targets out-of-distribution (OOD) generalization of distilled datasets. We study this problem through a widely adopted DD baseline of Distribution Matching (DM). We attribute the OOD vulnerability of DM to the entanglement of class-discriminative and domain-specific information within the compressed synthetic set, and propose Spectral Gradient Surgery (SGS) to disentangle the two. The key insight of SGS is that cross-domain agreement among domain-wise gradients in the spectral domain reveals which gradient components are shared across source domains-and are therefore class-discriminative-and which are domain-specific. Based on this observation, SGS augments the standard DM update with two complementary gradients: one that reinforces cross-domain shared components and another that explicitly promotes diversity within the distilled dataset. Extensive experiments on diverse-scale benchmarks demonstrate that SGS substantially improves OOD generalization while remaining plug-and-play compatible with existing DM methods.

Continual Knowledge Updating in LLM Systems: Learning Through Multi-Timescale Memory Dynamics

LLMs are trained once, then deployed into a world that never stops changing. External memory compensates for this, but most systems manage it explicitly rather than letting it adapt on its own. Biological memory works differently: coupled multi-timescale dynamics make new associations immediately usable, strengthen what repetition confirms, and let the rest fade. We argue that external memory should follow a similar principle. In Memini, this view takes the form of an associative memory that organizes knowledge as a directed graph. Each edge carries two coupled internal variables, one fast and one slow, following the Benna-Fusi model of synaptic consolidation. From this coupling, episodic sensitivity, gradual consolidation, and selective forgetting emerge as facets of a single mechanism, reframing external memory as a learning substrate that reorganizes through its own dynamics.

preprint2023arXiv

GraB: Finding Provably Better Data Permutations than Random Reshuffling

Random reshuffling, which randomly permutes the dataset each epoch, is widely adopted in model training because it yields faster convergence than with-replacement sampling. Recent studies indicate greedily chosen data orderings can further speed up convergence empirically, at the cost of using more computation and memory. However, greedy ordering lacks theoretical justification and has limited utility due to its non-trivial memory and computation overhead. In this paper, we first formulate an example-ordering framework named herding and answer affirmatively that SGD with herding converges at the rate $O(T^{-2/3})$ on smooth, non-convex objectives, faster than the $O(n^{1/3}T^{-2/3})$ obtained by random reshuffling, where $n$ denotes the number of data points and $T$ denotes the total number of iterations. To reduce the memory overhead, we leverage discrepancy minimization theory to propose an online Gradient Balancing algorithm (GraB) that enjoys the same rate as herding, while reducing the memory usage from $O(nd)$ to just $O(d)$ and computation from $O(n^2)$ to $O(n)$, where $d$ denotes the model dimension. We show empirically on applications including MNIST, CIFAR10, WikiText and GLUE that GraB can outperform random reshuffling in terms of both training and validation performance, and even outperform state-of-the-art greedy ordering while reducing memory usage over $100\times$.

preprint2015arXiv

Ordered Decompositional DAG Kernels Enhancements

In this paper, we show how the Ordered Decomposition DAGs (ODD) kernel framework, a framework that allows the definition of graph kernels from tree kernels, allows to easily define new state-of-the-art graph kernels. Here we consider a fast graph kernel based on the Subtree kernel (ST), and we propose various enhancements to increase its expressiveness. The proposed DAG kernel has the same worst-case complexity as the one based on ST, but an improved expressivity due to an augmented set of features. Moreover, we propose a novel weighting scheme for the features, which can be applied to other kernels of the ODD framework. These improvements allow the proposed kernels to improve on the classification performances of the ST-based kernel for several real-world datasets, reaching state-of-the-art performances.

AI-Guided Human-In-the-Loop Inverse Design of High Performance Engineering Structures

Inverse design tools such as Topology Optimization (TO) can achieve new levels of improvement for high-performance engineered structures. However, widespread use is hindered by high computational times and a black-box nature that inhibits user interaction. Human-in-the-loop TO approaches are emerging that integrate human intuition into the design generation process. However, these rely on the time-consuming bottleneck of iterative region selection for design modifications. To reduce the number of iterative trials, this contribution presents an AI co-pilot that uses machine learning to predict the user's preferred regions. The prediction model is configured as an image segmentation task with a U-Net architecture. It is trained on synthetic datasets where human preferences either identify the longest topological member or the most complex structural connection. The model successfully predicts plausible regions for modification and presents them to the user as AI recommendations. The human preference model demonstrates generalization across diverse and non-standard TO problems and exhibits emergent behavior outside the single-region selection training data. Demonstration examples

Fair inference on error-prone outcomes

Fair inference in supervised learning is an important and active area of research, yielding a range of useful methods to assess and account for fairness criteria when predicting ground truth targets. As shown in recent work, however, when target labels are error-prone, potential prediction unfairness can arise from measurement error. In this paper, we show that, when an error-prone proxy target is used, existing methods to assess and calibrate fairness criteria do not extend to the true target variable of interest. To remedy this problem, we suggest a framework resulting from the combination of two existing literatures: fair ML methods, such as those found in the counterfactual fairness literature on the one hand, and, on the other, measurement models found in the statistical literature. We discuss these approaches and their connection resulting in our framework. In a healthcare decision problem, we find that using a latent variable model to account for measurement error removes the unfairness detected previously.

Compressed Video Aggregator: Content-driven Module for Efficient Micro-Video Recommendation

We propose Compressed Video Aggregator (CVA), a lightweight micro-video recommendation module that decouples video information from preference learning. It aggregates frozen VFM embeddings, and uses latent reasoning without cross-attention projection, producing compact video embeddings for recommenders. Due to the redundancy in the frame count of the original benchmark and its overly coarse sampling, we used titles to re-select key frames based on CLIP. Experiments on MicroLens and Short-Video show consistent gains with orders-of-magnitude reductions in training time and GPU memory, and re-selected frames can further enhance the performance of all methods, including CVA. Furthermore, we also discussed the impact of several scenarios involving erroneous titles on our method. Code will be released soon.

On the Consistency of Optimal Bayesian Feature Selection in the Presence of Correlations

Optimal Bayesian feature selection (OBFS) is a multivariate supervised screening method designed from the ground up for biomarker discovery. In this work, we prove that Gaussian OBFS is strongly consistent under mild conditions, and provide rates of convergence for key posteriors in the framework. These results are of enormous importance, since they identify precisely what features are selected by OBFS asymptotically, characterize the relative rates of convergence for posteriors on different types of features, provide conditions that guarantee convergence, justify the use of OBFS when its internal assumptions are invalid, and set the stage for understanding the asymptotic behavior of other algorithms based on the OBFS framework.

preprint2015arXiv

Bootstrapped Thompson Sampling and Deep Exploration

This technical note presents a new approach to carrying out the kind of exploration achieved by Thompson sampling, but without explicitly maintaining or sampling from posterior distributions. The approach is based on a bootstrap technique that uses a combination of observed and artificially generated data. The latter serves to induce a prior distribution which, as we will demonstrate, is critical to effective exploration. We explain how the approach can be applied to multi-armed bandit and reinforcement learning problems and how it relates to Thompson sampling. The approach is particularly well-suited for contexts in which exploration is coupled with deep learning, since in these settings, maintaining or generating samples from a posterior distribution becomes computationally infeasible.

preprint2017arXiv

Fast Kronecker product kernel methods via generalized vec trick

Kronecker product kernel provides the standard approach in the kernel methods literature for learning from graph data, where edges are labeled and both start and end vertices have their own feature representations. The methods allow generalization to such new edges, whose start and end vertices do not appear in the training data, a setting known as zero-shot or zero-data learning. Such a setting occurs in numerous applications, including drug-target interaction prediction, collaborative filtering and information retrieval. Efficient training algorithms based on the so-called vec trick, that makes use of the special structure of the Kronecker product, are known for the case where the training data is a complete bipartite graph. In this work we generalize these results to non-complete training graphs. This allows us to derive a general framework for training Kronecker product kernel methods, as specific examples we implement Kronecker ridge regression and support vector machine algorithms. Experimental results demonstrate that the proposed approach leads to accurate models, while allowing order of magnitude improvements in training and prediction time.