Large language models (LLMs) have significantly transformed natural language understanding and generation, but they raise privacy concerns due to potential exposure of sensitive information. Studies have highlighted the risk of information leakage, where adversaries can extract sensitive information embedded in the prompts. In this work, we introduce a novel private prediction framework for generating high-quality synthetic text with strong privacy guarantees. Our approach leverages the Differential Privacy (DP) framework to ensure worst-case theoretical bounds on information leakage without requiring any fine-tuning of the underlying models. The proposed method performs inference on private records and aggregates the resulting per-token output distributions. This enables the generation of longer and coherent synthetic text while maintaining privacy guarantees. Additionally, we propose a simple blending operation that combines private and public inference to further enhance utility. Empirical evaluations demonstrate that our approach outperforms previous state-of-the-art methods on in-context-learning (ICL) tasks, making it a promising direction for privacy-preserving text generation while maintaining high utility.

We propose a novel inference-time out-ofdomain (OOD) detection algorithm for specialized large language models (LLMs).Despite achieving state-of-the-art performance on in-domain tasks through fine-tuning, specialized LLMs remain vulnerable to incorrect or unreliable outputs when presented with OOD inputs, posing risks in critical applications.Our method leverages the Inductive Conformal Anomaly Detection (ICAD) framework, using a new non-conformity measure based on the model's dropout tolerance.Motivated by recent findings on polysemanticity and redundancy in LLMs, we hypothesize that in-domain inputs exhibit higher dropout tolerance than OOD inputs.We aggregate dropout tolerance across multiple layers via a valid ensemble approach, improving detection while maintaining theoretical false alarm bounds from ICAD.Experiments with medical-specialized LLMs show that our approach detects OOD inputs better than baseline methods, with AUROC improvements of 2% to 37% when treating OOD datapoints as positives and in-domain test datapoints as negatives.

Despite their widespread use, large language models (LLMs) are known to hallucinate incorrect information and be poorly calibrated. This makes the uncertainty quantification of these models of critical importance, especially in high-stakes domains, such as autonomy and healthcare. Prior work has made Bayesian deep learning-based approaches to this problem more tractable by performing inference over the low-rank adaptation (LoRA) parameters of a fine-tuned model. While effective, these approaches struggle to scale to larger LLMs due to requiring further additional parameters compared to LoRA. In this work we present $\textbf{Scala}$ble $\textbf{B}$ayesian $\textbf{L}$ow-Rank Adaptation via Stochastic Variational Subspace Inference (ScalaBL). We perform Bayesian inference in an $r$-dimensional subspace, for LoRA rank $r$. By repurposing the LoRA parameters as projection matrices, we are able to map samples from this subspace into the full weight space of the LLM. This allows us to learn all the parameters of our approach using stochastic variational inference. Despite the low dimensionality of our subspace, we are able to achieve competitive performance with state-of-the-art approaches while only requiring ${\sim}1000$ additional parameters. Furthermore, it allows us to scale up to the largest Bayesian LLM to date, with four times as a many base parameters as prior work.

We address the problem of detecting out-of-context (OOC) objects in a scene. Given an image, we aim to detect whether the image has objects that are not present in their usual context and localize such OOC objects. Existing approaches for OOC detection rely on defining the common context in terms of the manually constructed features, such as the co-occurrence of objects, spatial relations between objects, and shape and size of the objects, and then learning such context for a given dataset. But context is often nu-anced ranging from very common to very surprising. Further, learned context from specific datasets may not be generalized as datasets may not truly represent the human notion of what is in context. Motivated by the success of large language models and more generally, foundation models (FMs) in common sense reasoning, we investigate the FM's ability to capture a more generalized notion of context. We find that a pre-trained FM, such as GPT-4, provides a more nuanced notion of OOC and enables zero-shot OOC detection when coupled with other pre-trained FMs for caption generation such as BLIP-2, and image in-painting with Sta-ble Diffusion 2.0. Our approach does not need any dataset-specific training. We demonstrate the efficacy of our approach on two OOC object detection datasets, achieving 90.8% zero-shot accuracy on the MIT-OOC dataset and 87.26% on the IJCAI22-COCO-OOC dataset.

Uncertainty in the predictions of learning-enabled components hinders their deployment in safety-critical cyber-physical systems (CPS). A shift from the training distribution of a learning-enabled component (LEC) is one source of uncertainty in the LEC’s predictions. Detection of this shift or out-of-distribution (OOD) detection on individual datapoints has therefore gained attention recently. But in many applications, inputs to CPS form a temporal sequence. Existing techniques for OOD detection in time-series data for CPS either do not exploit temporal relationships in the sequence or do not provide any guarantees on detection. We propose using deviation from the in-distribution temporal equivariance as the non-conformity measure in conformal anomaly detection framework for OOD detection in time-series data for CPS. Computing independent predictions from multiple conformal detectors based on the proposed measure and combining these predictions by Fisher’s method leads to the proposed detector CODiT with bounded false alarms. CODiT performs OOD detection on fixed-length windows of consecutive time-series datapoints by using Fisher value of the input window. We further propose performing OOD detection on real-time time-series traces of variable lengths with bounded false alarms. This can be done by using CODiT to compute Fisher values of the sliding windows in the input trace and combining these values by a merging function. Merging functions such as Harmonic Mean, Arithmetic Mean, Geometric Mean, Bonferroni Method, and so on, can be used to combine Fisher values of the sliding windows in the input trace, and the combined value can be used for OOD detection on the trace with bounded false alarm rate guarantees. We illustrate the efficacy of CODiT by achieving state-of-the-art results in two case studies for OOD detection on fixed-length windows. The first one is on an autonomous driving system with perception (or vision) LEC. The second case study is on a medical CPS for walking pattern or GAIT analysis where physiological (non-vision) data is collected with force-sensitive resistors attached to the subject’s body. For OOD detection on variable length traces, we consider the same case studies on the autonomous driving system and medical CPS for GAIT analysis. We report our results with four merging functions on the Fisher values computed by CODiT on the sliding windows of the input trace. We also compare the false alarm rate guarantees by these four merging functions in the autonomous driving system case study. Code, data, and trained models are available at https://github.com/kaustubhsridhar/time-series-OOD .

Incorporating learning based components in the current state-of-the-art cyber-physical systems (CPS) has been a challenge due to the brittleness of the underlying deep neural networks. On the bright side, if executed correctly with safety guarantees, this has the ability to revolutionize domains like autonomous systems, medicine, and other safety-critical domains. This is because it would allow system designers to use high-dimensional outputs from sensors like camera and LiDAR. The trepidation in deploying systems with vision and LiDAR components comes from incidents of catastrophic failures in the real world. Recent reports of self-driving cars running into difficult to handle scenarios is ingrained in the software components which handle such sensor inputs. The ability to handle such high-dimensional signals is due to the explosion of algorithms which use deep neural networks. Sadly, the reason behind the safety issues is also due to deep neural networks themselves. The pitfalls occur due to possible over-fitting and lack of awareness about the blind spots induced by the training distribution. Ideally, system designers would wish to cover as many scenarios during training as possible. However, achieving a meaningful coverage is impossible. This naturally leads to the following question: is it feasible to flag out-of-distribution (OOD) samples without causing too many false alarms? Such an OOD detector should be executable in a fashion that is computationally efficient. This is because OOD detectors often are executed as frequently as the sensors are sampled. Our aim in this article is to build an effective anomaly detector. To this end, we propose the idea of a memory bank to cache data samples which are representative enough to cover most of the in-distribution data. The similarity with respect to such samples can be a measure of familiarity of the test input. This is made possible by an appropriate choice of distance function tailored to the type of sensor we are interested in. Additionally, we adapt conformal anomaly detection framework to capture the distribution shifts with a guarantee of false alarm rate. We report the performance of our technique on two challenging scenarios: a self-driving car setting implemented inside the simulator CARLA with image inputs and autonomous racing car navigation setting with LiDAR inputs. From the experiments, it is clear that a deviation from the in-distribution setting can potentially lead to unsafe behavior. It should be noted that not all OOD inputs lead to precarious situations in practice, but staying in-distribution is akin to staying within a safety bubble and predictable behavior. An added benefit of our memory-based approach is that the OOD detector produces interpretable feedback for a human designer. This is of utmost importance since it recommends a potential fix for the situation as well. In other competing approaches, such feedback is difficult to obtain due to reliance on techniques which use variational autoencoders.

Machine learning methods such as deep neural networks (DNNs), despite their success across different domains, are known to often generate incorrect predictions with high confidence on inputs outside their training distribution. The deployment of DNNs in safety-critical domains requires detection of out-of-distribution (OOD) data so that DNNs can abstain from making predictions on those. A number of methods have been recently developed for OOD detection, but there is still room for improvement. We propose the new method iDECODe, leveraging in-distribution equivariance for conformal OOD detection. It relies on a novel base non-conformity measure and a new aggregation method, used in the inductive conformal anomaly detection framework, thereby guaranteeing a bounded false detection rate. We demonstrate the efficacy of iDECODe by experiments on image and audio datasets, obtaining state-of-the-art results. We also show that iDECODe can detect adversarial examples. Code, pre-trained models, and data are available at https://github.com/ramneetk/iDECODe.

Signal Temporal Logic (STL) is a prominent specification formalism for real-time systems, and monitoring these specifications, specially when (for different reasons such as learning) behavior of systems can change over time, is quite important. There are three main challenges in this area: (1) full observation of system state is not possible due to noise or nuisance parameters, (2) the whole execution is not available during the monitoring, and (3) computational complexity of monitoring continuous time signals is very high. Although, each of these challenges has been addressed by different works, to the best of our knowledge, no one has addressed them all together. In this paper, we show how to extend any parameter invariant test procedure for single points in time to a parameter invariant test procedure for efficiently monitoring continuous time executions of a system against STL properties. We also show, how to extend probabilistic error guarantee of the input test procedure to a probabilistic error guarantee for the constructed test procedure.