We introduce SelfCite, a novel self-supervised approach that aligns LLMs to generate high-quality, fine-grained, sentence-level citations for the statements in their generated responses. Instead of only relying on costly and labor-intensive annotations, SelfCite leverages a reward signal provided by the LLM itself through context ablation: If a citation is necessary, removing the cited text from the context should prevent the same response; if sufficient, retaining the cited text alone should preserve the same response. This reward can guide the inference-time best-of-N sampling strategy to improve citation quality significantly, as well as be used in preference optimization to directly fine-tune the models for generating better citations. The effectiveness of SelfCite is demonstrated by increasing citation F1 up to 5.3 points on the LongBench-Cite benchmark across five long-form question answering tasks.

We present a theoretical explanation of the ``grokking'' phenomenon, where a model generalizes long after overfitting,for the originally-studied problem of modular addition. First, we show that early in gradient descent, when the ``kernel regime'' approximately holds, no permutation-equivariant model can achieve small population error on modular addition unless it sees at least a constant fraction of all possible data points. Eventually, however, models escape the kernel regime. We show that two-layer quadratic networks that achieve zero training loss with bounded $\ell_{\infty}$ norm generalize well with substantially fewer training points, and further show such networks exist and can be found by gradient descent with small $\ell_{\infty}$ regularization. We further provide empirical evidence that these networks as well as simple Transformers, leave the kernel regime only after initially overfitting. Taken together, our results strongly support the case for grokking as a consequence of the transition from kernel-like behavior to limiting behavior of gradient descent on deep networks.

This paper presents a self-supervised learning method to safely learn a motion planner for ground robots to navigate environments with dense and dynamic obstacles. When facing highly-cluttered, fast-moving, hard-to-predict obstacles, classical motion planners may not be able to keep up with limited onboard computation. For learning-based planners, high-quality demonstrations are difficult to acquire for imitation learning while reinforcement learning becomes inefficient due to the high probability of collision during exploration. To safely and efficiently provide training data, the Learning from Hallucination (LfH) approaches synthesize difficult navigation environments based on past successful navigation experiences in relatively easy or completely open ones, but unfortunately cannot address dynamic obstacles. In our new Dynamic Learning from Learned Hallucination (Dyna-LfLH), we design and learn a novel latent distribution and sample dynamic obstacles from it, so the generated training data can be used to learn a motion planner to navigate in dynamic environments. Dyna-LfLH is evaluated on a ground robot in both simulated and physical environments and achieves up to 25% better success rate compared to baselines.

To equip Convolutional Neural Networks (CNNs) with explainability, it is essential to interpret how opaque models take specific decisions, understand what causes the errors, improve the architecture design, and identify unethical biases in the classifiers. This paper introduces ADVISE, a new explainability method that quantifies and leverages the relevance of each unit of the feature map to provide better visual explanations. To this end, we propose using adaptive bandwidth kernel density estimation to assign a relevance score to each unit of the feature map with respect to the predicted class. We also propose an evaluation protocol to quantitatively assess the visual explainability of CNN models. We extensively evaluate our idea in the image classification task using AlexNet, VGG16, ResNet50, and Xception pretrained on ImageNet. We compare ADVISE with the state-of-the-art visual explainable methods and show that the proposed method outperforms competing approaches in quantifying feature-relevance and visual explainability while maintaining competitive time complexity. Our experiments further show that ADVISE fulfils the sensitivity and implementation independence axioms while passing the sanity checks. The implementation is accessible for reproducibility purposes on https://github.com/dehshibi/ADVISE.

Deep reinforcement learning algorithms require large and diverse datasets in order to learn successful policies for perception-based mobile navigation. However, gathering such datasets with a single robot can be prohibitively expensive. Collecting data with multiple different robotic platforms with possibly different dynamics is a more scalable approach to large-scale data collection. But how can deep reinforcement learning algorithms leverage such heterogeneous datasets? In this work, we propose a deep reinforcement learning algorithm with hierarchically integrated models (HInt). At training time, HInt learns separate perception and dynamics models, and at test time, HInt integrates the two models in a hierarchical manner and plans actions with the integrated model. This method of planning with hierarchically integrated models allows the algorithm to train on datasets gathered by a variety of different platforms, while respecting the physical capabilities of the deployment robot at test time. Our mobile navigation experiments show that HInt outperforms conventional hierarchical policies and single-source approaches.

Statistical Hypothesis Testing (SHT) is a class of inference methods whereby one makes use of empirical data to test a hypothesis and often emit a judgment about whether to reject it or not. In this paper we focus on the logical aspect of this strategy, which is largely independent of the adopted school of thought, at least within the various frequentist approaches. We identify SHT as taking the form of an unsound argument from Modus Tollens in classical logic, and, in order to rescue SHT from this difficulty, we propose that it can instead be grounded in t-norm based fuzzy logics. We reformulate the frequentists' SHT logic by making use of a fuzzy extension of modus Tollens to develop a model of truth valuation for its premises. Importantly, we show that it is possible to preserve the soundness of Modus Tollens by exploring the various conventions involved with constructing fuzzy negations and fuzzy implications (namely, the S and R conventions). We find that under the S convention, it is possible to conduct the Modus Tollens inference argument using Zadeh's compositional extension and any possible t-norm. Under the R convention we find that this is not necessarily the case, but that by mixing R-implication with S-negation we can salvage the product t-norm, for example. In conclusion, we have shown that fuzzy logic is a legitimate framework to discuss and address the difficulties plaguing frequentist interpretations of SHT.

For dynamic manipulation of flexible objects, we propose an acquisition method of a flexible object motion equation model using a deep neural network and a control method to realize a target state by calculating an optimized time-series joint torque command. By using the proposed method, any physics model of a target object is not needed, and the object can be controlled as intended. We applied this method to manipulations of a rigid object, a flexible object with and without environmental contact, and a cloth, and verified its effectiveness.