IoT is rapidly expanding from traditional small-scale (0–100m) applications like smart homes and large-scale (1–5km) applications like Microsoft’s FarmBeats to emerging mesoscale (0.1–1.5km) applications such as smart-grid NANs and peer-to-peer energy trading in smart homes. These applications demand high throughput and low latency but currently lack dedicated radio technologies. Our qualitative analysis identified Zigbee and LoRa as promising candidates. Further quantitative analysis revealed that a multi-radio architecture combining these radios achieves the best throughput. However, within the 500–1200m range, termed the gray region, it is unpredictable which radio offers higher throughput at any given moment. To address this, we developed MARS, a Multi-radio Architecture with Radio Selection, powered by TAO-optimized decision trees that select the high-throughput radio at the time of transmission. These decision trees require instantaneous path quality estimates, but traditional multi-hop Zigbee networks cannot provide these promptly due to propagation and queuing delays. We overcome this challenge by introducing Decision Tree-based updates to instantaneously estimate end-to-end path quality. Large-scale, real-world experiments with MARS demonstrated average throughput gains of 48.2% and 49.79% at two different locations.

Large Language Models (LLMs) are vulnerable to **jailbreaking attacks** that lead to generation of inappropriate or harmful content. Manual red-teaming requires a time-consuming search for adversarial prompts, whereas automatic adversarial prompt generation often leads to semantically meaningless attacks that do not scale well. In this paper, we present a novel method that uses another LLM, called **AdvPrompter**, to generate human-readable adversarial prompts in seconds. AdvPrompter, which is trained using an alternating optimization algorithm, generates suffixes that veil the input instruction without changing its meaning, such that the TargetLLM is lured to give a harmful response. Experimental results on popular open source TargetLLM show highly competitive results on the AdvBench and HarmBench datasets, that also transfer to closed-source black-box LLMs. We also show that training on adversarial suffixes generated by AdvPrompter is a promising strategy for improving the robustness of LLMs to jailbreaking attacks.

Autonomous UI agents powered by AI have tremendous potential to boost human productivity by automating routine tasks such as filing taxes and paying bills. However, a major challenge in unlocking their full potential is security, which is exacerbated by the agent's ability to take action on their user's behalf. Existing tests for prompt injections in web agents either over-simplify the threat by testing unrealistic scenarios or giving the attacker too much power, or look at single-step isolated tasks. To more accurately measure progress for secure web agents, we introduce WASP – a new publicly available benchmark for end-to-end evaluation of Web Agent Security against Prompt Injection attacks. Evaluating with WASP shows that even top-tier AI models, including those with advanced reasoning capabilities, can be deceived by simple, low-effort human-written injections in very realistic scenarios. Our end-to-end evaluation reveals a previously unobserved insight: while attacks partially succeed in up to 86% of the case, even state-of-the-art agents often struggle to fully complete the attacker goals – highlighting the current state of security by incompetence. Code and data are available at https://github.com/facebookresearch/wasp.

Despite progress in language model (LM) capabilities, evaluations have thus far focused on models' performance on tasks that humans have previously solved, including in programming (SWE-Bench) and mathematics (FrontierMath). We therefore propose testing models' ability to design and implement algorithms in an open-ended benchmark: We task LMs with writing code that efficiently solves computationally challenging problems in computer science, physics, and mathematics. Our AlgoTune benchmark consists of 120 tasks collected from domain experts and a framework for validating and timing LM-synthesized solution code, which is compared to reference implementations from popular open-source packages.In addition, we develop a baseline LM agent, AlgoTuner, and evaluate its performance across a suite of frontier models.AlgoTuner achieves an average 1.58x speedup against reference solvers, including methods from packages such as SciPy, scikit-learn and CVXPY.However, we find that current models fail to discover algorithmic innovations, instead preferring surface-level optimizations. We hope that AlgoTune catalyzes the development of LM agents exhibiting creative problem solving beyond state-of-the-art human performance.

Autonomous AI agents that can follow instructions and perform complex multi-step tasks have tremendous potential to boost human productivity. However, to perform many of these tasks, the agents need access to personal information from their users, raising the question of whether they are capable of using it appropriately. In this work, we introduce a new benchmark AgentDAM that measures if AI web-navigation agents follow the privacy principle of "data minimization". For the purposes of our benchmark, data minimization means that the agent uses a piece of potentially sensitive information only if it is "necessary" to complete a particular task. Our benchmark simulates realistic web interaction scenarios end-to-end and is adaptable to all existing web navigation agents. We use AgentDAM to evaluate how well AI agents built on top of GPT-4, Llama-3 and Claude can limit processing of potentially private information, and show that they are prone to inadvertent use of unnecessary sensitive information. We also propose a prompting-based defense that reduces information leakage, and demonstrate that our end-to-end benchmarking provides a more realistic measure than probing LLMs about privacy. Our results highlight that further research is needed to develop AI agents that can prioritize data minimization at inference time. We open source our benchmark at: https://github.com/facebookresearch/ai-agent-privacy

Deep generative models like GAN and VAE have shown impressive results in generating unconstrained objects like images. However, many design settings arising in industrial design, material science, computer graphics and more require that the generated objects satisfy hard combinatorial constraints or meet objectives in addition to modeling a data distribution. To address this, we propose GenCO, a generative framework that guarantees constraint satisfaction throughout training by leveraging differentiable combinatorial solvers to enforce feasibility. GenCO imposes the generative loss on provably feasible solutions rather than intermediate soft solutions, meaning that the deep generative network can focus on ensuring the generated objects match the data distribution without having to also capture feasibility. This shift enables practitioners to enforce hard constraints on the generated outputs during end-to-end training, enabling assessments of their feasibility and introducing additional combinatorial loss components to deep generative training. We demonstrate the effectiveness of our approach on a variety of generative combinatorial tasks, including game level generation, map creation for path planning, and photonic device design, consistently demonstrating its capability to yield diverse, high-quality solutions that verifiably adhere to user-specified combinatorial properties.

Important applications such as fraud or spam detection or churn prediction involve binary classification problems where the datasets are imbalanced and the cost of false positives greatly differs from the cost of false negatives. We focus on classification trees, in particular oblique trees, which subsume both the traditional axis-aligned trees and logistic regression, but are more accurate than both while providing interpretable models. Rather than using ROC curves, we advocate a loss based on minimizing the false negatives subject to a maximum false positive rate, which we prove to be equivalent to minimizing a weighted 0/1 loss. This yields a curve of classifiers that provably dominates the ROC curve, but is hard to optimize due to the 0/1 loss. We give the first algorithm that can iteratively update the tree parameters globally so that the weighted 0/1 loss decreases monotonically. Experiments on various datasets with class imbalance or class costs show this indeed dominates ROC-based classifiers and significantly improves over previous approaches to learn trees based on weighted purity criteria or over- or undersampling.

Mixed integer linear programs (MILP) are flexible and powerful tools for modeling and solving many difficult real-world combinatorial optimization problems. In this paper, we propose a novel machine learning (ML)-based framework ConPaS that learns to predict solutions to MILPs with contrastive learning. For training, we collect high-quality solutions as positive samples. We also collect low-quality or infeasible solutions as negative samples using novel optimization-based or sampling approaches. We then learn to make discriminative predictions by contrasting the positive and negative samples. During testing, we predict and fix the assignments for a subset of integer variables and then solve the resulting reduced MILP to find high-quality solutions. Empirically, ConPaS achieves state-of-the-art results compared to other ML-based approaches in terms of the quality of and the speed at which solutions are found.

Decision forests are among the most accurate models in machine learning. This is remarkable given that the way they are trained is highly heuristic: neither the individual trees nor the overall forest optimize any well-defined loss. While diversity mechanisms such as bagging or boosting have been until now critical in the success of forests, we think that a better optimization should lead to better forests—ideally eliminating any need for an ensembling heuristic. However, unlike for most other models, such as neural networks, optimizing forests or trees is not easy, because they define a non-differentiable function. We show, for the first time, that it is possible to learn a forest by optimizing a desirable loss and regularization jointly over all its trees and parameters. Our algorithm, Forest Alternating Optimization, is based on defining a forest as a parametric model with a fixed number of trees and structure (rather than adding trees indefinitely as in bagging or boosting). It then iteratively updates each tree in alternation so that the objective function decreases monotonically. The algorithm is so effective at optimizing that it easily overfits, but this can be corrected by averaging. The result is a forest that consistently exceeds the accuracy of the state-of-the-art while using fewer, smaller trees.

Recent works in learning-integrated optimization have shown promise in settings where the optimization problem is only partially observed or where general-purpose optimizers perform poorly without expert tuning. By learning an optimizer $\mathbf{g}$ to tackle these challenging problems with $f$ as the objective, the optimization process can be substantially accelerated by leveraging past experience. The optimizer can be trained with supervision from known optimal solutions or implicitly by optimizing the compound function $f\circ \mathbf{g}$. The implicit approach may not require optimal solutions as labels and is capable of handling problem uncertainty; however, it is slow to train and deploy due to frequent calls to optimizer $\mathbf{g}$ during both training and testing. The training is further challenged by sparse gradients of $\mathbf{g}$, especially for combinatorial solvers. To address these challenges, we propose using a smooth and learnable **Landscape Surrogate** $\mathcal{M}$ as a replacement for $f\circ \mathbf{g}$. This surrogate, learnable by neural networks, can be computed faster than the solver $\mathbf{g}$, provides dense and smooth gradients during training, can generalize to unseen optimization problems, and is efficiently learned via alternating optimization. We test our approach on both synthetic problems, including shortest path and multidimensional knapsack, and real-world problems such as portfolio optimization, achieving comparable or superior objective values compared to state-of-the-art baselines while reducing the number of calls to $\mathbf{g}$. Notably, our approach outperforms existing methods for computationally expensive high-dimensional problems.

Model interpretability is a topic of renewed interest given today’s widespread practical use of machine learning, and the need to trust or understand automated predictions. We consider the problem of optimally learning interpretable out-of-sample mappings for nonlinear embedding methods such as $t$-SNE. We argue for the use of sparse oblique decision trees because they strike a good tradeoff between accuracy and interpretability which can be controlled via a hyperparameter, thus allowing one to achieve a model with a desired explanatory complexity. The resulting optimization problem is difficult because decision trees are not differentiable. By using an equivalent formulation of the problem, we give an algorithm that can learn such a tree for any given nonlinear embedding objective. We illustrate experimentally how the resulting trees provide insights into the data beyond what a simple 2D visualization of the embedding does.

Acoustic events have a hierarchical structure analogous to a tree (or a directed acyclic graph). In this work, we propose a structure-aware semi-supervised learning framework for acoustic event classification (AEC). Our hypothesis is that the audio label structure contains useful information that is not available in audios and plain tags. We show that by organizing audio representations with a human-curated tree ontology, we can improve the quality of the learned audio representations for downstream AEC tasks. We use consistency training to use large amounts of unlabeled data for structured representation manifold learning. Experimental results indicate that our framework learns high quality representations which enable us to achieve comparable performance in discriminative tasks as fully supervised baselines. Moreover, our framework can better handle audios with unseen tags by confidently assigning a super-category (internal node like "animal" in Fig. 1) tag to the audio.

Boosting, one of the most effective machine learning frameworks, has attracted an enduring interest since its introduction 30 years ago. The majority of boosting methods use trees as base learner and, while much work has focused on theoretical and empirical variations of boosting, there has been surprisingly little progress on the tree learning procedure itself. To this day, each individual tree is typically axis-aligned (which is ill-suited to model correlations and results in relatively weak classifiers), and is learned using a greedy divide-and-conquer approach such as CART or C5.0, which produces suboptimal trees. We show we can improve boosted forests drastically by making each tree a much stronger classifier. We do this by using sparse oblique trees, which are far more powerful than axis-aligned ones, and by optimizing them using “tree alternating optimization” (TAO), suitably modified to handle the base learner optimization problem dictated by the boosting framework. Focusing on two versions of AdaBoost, we show that the resulting forests not only are consistently and considerably more accurate than random forests or gradient boosting, but that they use a very small number of trees and a comparable number of parameters.

Semi-supervised learning seeks to learn a machine learning model when only a small amount of the available data is labeled. The most widespread approach uses a graph prior, which encourages similar instances to have similar predictions. This has been very successful with models ranging from kernel machines to neural networks, but has remained inapplicable to decision trees, for which the optimization problem is much harder. We solve this based on a reformulation of the problem which requires iteratively solving two simpler problems: a supervised tree learning problem, which can be solved by the Tree Alternating Optimization algorithm; and a label smoothing problem, which can be solved through a sparse linear system. The algorithm is scalable and highly effective even with very few labeled instances, and makes it possible to learn accurate, interpretable models based on decision trees in such situations.

Classification problems having thousands or more classes naturally occur in NLP, for example language models or document classification.A softmax or one-vs-all classifier naturally handles many classes, but it is very slow at inference time, because every class score must be calculated to find the top class.We propose the "softmax tree", consisting of a binary tree having sparse hyperplanes at the decision nodes (which make hard, not soft, decisions) and small softmax classifiers at the leaves.This is much faster at inference because the input instance follows a single path to a leaf (whose length is logarithmic on the number of leaves) and the softmax classifier at each leaf operates on a small subset of the classes.Although learning accurate tree-based models has proven difficult in the past, we are able to overcome this by using a variation of a recent algorithm, tree alternating optimization (TAO).Compared to a softmax and other classifiers, the resulting softmax trees are both more accurate in prediction and faster in inference, as shown in NLP problems having from one thousand to one hundred thousand classes.

Much of the success of deep learning is due to choosing good neural net architectures and being able to train them effectively. A type of architecture that has been long sought is one that combines decision trees and neural nets. This is straightforward if the tree makes soft decisions (i.e., an input instance follows all paths in the tree with different probabilities), because the model is differentiable. However, the optimization is much harder if the tree makes hard decisions, but this produces an architecture that is much faster at inference, since an instance follows a single path in the tree. We show that it is possible to train such architectures, with guaranteed monotonic decrease of the loss, and demonstrate it by learning trees with linear decision nodes and deep nets at the leaves. The resulting architecture improves state-of-the-art deep nets, by achieving comparable or lower classification error but with fewer parameters and faster inference time. In particular, we show that, rather than improving a ResNet by making it deeper, it is better to construct a tree of small ResNets. The resulting tree-net hybrid is also more interpretable.

The widespread deployment of deep nets in practical applications has lead to a growing desire to understand how and why such black-box methods perform prediction. Much work has focused on understanding what part of the input pattern (an image, say) is responsible for a particular class being predicted, and how the input may be manipulated to predict a different class. We focus instead on understanding what internal features computed by the neural net are responsible for a particular class. We achieve this by mimicking part of the net with a decision tree having sparse weight vectors at the nodes. We are able to learn trees that are both highly accurate and interpretable, so they can provide insights into the deep net black box. Further, we show we can easily manipulate the neural net features in order to make the net predict, or not predict, a given class, thus showing that it is possible to carry out adversarial attacks at the level of the features. We demonstrate this robustly in MNIST and ImageNet with LeNet5 and VGG networks.

We propose a new type of ensemble method that is specially designed for neural nets, and which produces surprising improvements in accuracy at a very small cost, without requiring to train a new neural net. The idea is to concatenate the output activations of internal layers of the neural net into an “ensemble feature vector”, and train on this a decision tree to predict the class labels while also doing feature selection. For this to succeed we rely on a recently proposed algorithm to train decision trees - Tree Alternating Optimization. This simple procedure consistently improves over simply ensembling the nets in the traditional way, achieving relative error decreases of well over 10% of the original nets on the well known image classification benchmarks. As a subproduct, we also can obtain an architecture consisting of a neural net feature extraction followed by a tree classifier that is faster and more compact than the original net.

Decision tree boosting is considered as an important and widely recognized method in image classification, despite dominance of the deep learning based approaches in this area. Provided with good image features, it can produce a powerful model with unique properties, such as strong predictive power, scalability, interpretability, etc. In this paper, we propose a novel tree boosting framework which capitalizes on the idea of using shallow, sparse and yet powerful oblique decision trees (trained with recently proposed Tree Alternating optimization algorithm) as the base learners. We empirically show that the resulting model achieves better or comparable performance (both in terms of accuracy and model size) against established boosting algorithms such as gradient boosting or AdaBoost in number of benchmarks. Further, we show that such trees can directly and efficiently handle multiclass problems without using one-vs-all strategy employed by most of the practical boosting implementations.

Regression forests (ensembles of trees) are considered as the leading off-the-shelf method for regression. One of the main approaches of constructing such forests is based on boosting. However, majority of the current boosting implementations employ an axis-aligned tree as a base learner, where each decision node tests for a single feature. Moreover, such trees are usually trained by greedy top-down algorithms such as CART which is shown to be suboptimal. We instead use oblique trees, where each decision node tests for a linear combination of features and train them with the recently proposed non-greedy tree learning method-Tree Alternating Optimization (TAO). We embed the TAO algorithm into the boosting framework and show its effectiveness in the regression setting. We show that it produces much better forests than other types of tree ensembling methods in terms of error, model size and inference time. The result has an immense practical impact on various applications such as in signal processing, data mining, computer vision, etc.

Learning decision trees is a difficult optimization problem: nonconvex, nondifferentiable and over a huge number of tree structures. The dominant paradigm in practice, established in the 1980s, are axis-aligned trees trained with a greedy recursive partitioning algorithm such as CART or C5.0. Several non-greedy optimization algorithms have been proposed recently, which optimize all the nodes' parameters jointly, and we compare experimentally some of them in a range of classification and regression datasets, in terms of accuracy, training time and tree size. The non-greedy algorithms do not improve over CART significantly with one exception, tree alternating optimization (TAO). TAO scales to large datasets and produces axis-aligned and especially oblique trees that consistently outperform all other algorithms, often by a large margin. TAO makes oblique trees preferable to axis-aligned ones in many cases, since they are much more accurate while remaining small and interpretable. This suggests a change in paradigm in practical applications of decision trees.

Regression forests, based on ensemble approaches such as bagging or boosting, have long been recognized as the leading off-the-shelf method for regression. However, forests rely on a greedy top-down procedure such as CART to learn each tree. We extend a recent algorithm for learning classification trees, Tree Alternating Optimization (TAO), to the regression case, and use it with bagging to construct regression forests of oblique trees, having hyperplane splits at the decision nodes. In a wide range of datasets, we show that the resulting forests exceed the accuracy of state-of-the-art algorithms such as random forests, AdaBoost or gradient boosting, often considerably, while yielding forests that have usually fewer and shallower trees and hence fewer parameters and faster inference overall. This result has an immense practical impact and advocates for the power of optimization in ensemble learning.