Value alignment, which aims to ensure that large language models (LLMs) and other AI agents behave in accordance with human values, is critical for ensuring safety and trustworthiness of these systems. A key component of value alignment is the modeling of human preferences as a representation of human values. In this paper, we investigate the robustness of value alignment by examining the sensitivity of preference models. Specifically, we ask: how do changes in the probabilities of some preferences affect the predictions of these models for other preferences? To answer this question, we theoretically analyze the robustness of widely used preference models by examining their sensitivities to minor changes in preferences they model. Our findings reveal that, in the Bradley-Terry and the Placket-Luce model, the probability of a preference can change significantly as other preferences change, especially when these preferences are dominant (i.e., with probabilities near zero or one). We identify specific conditions where this sensitivity becomes significant for these models and discuss the practical implications for the robustness and safety of value alignment in AI systems.

Human-Object Interaction (HOI) Detection constitutes an important aspect of human-centric scene understanding, which requires precise object detection and interaction recognition. Despite increasing advancement in detection, recognizing subtle and intricate interactions remains challenging. Recent methods have endeavored to leverage the rich semantic representation from pretrained CLIP, yet fail to efficiently capture finer-grained spatial features that are highly informative for interaction discrimination. In this work, instead of solely using representations from CLIP, we fill the gap by proposing a spatial adapter that efficiently utilizes the multi-scale spatial information in the pretrained detector. This leads to a bilateral adaptation that mutually produces complementary features. To further improve interaction recognition under occlusion, which is common in crowded scenarios, we propose an Occluded Part Extrapolation module that guides the model to recover the spatial details from manually occluded feature maps. Moreover, we design a Conditional Contextual Mining module that further mines informative contextual clues from the spatial features via a tailored cross-attention mechanism. Extensive experiments on V-COCO and HICO-DET benchmarks demonstrate that our method significantly outperforms prior art on both standard and zero-shot settings, resulting in new state-of-the-art performance. Additional ablation studies further validate the effectiveness of each component in our method.

Successful image classification requires a discriminative representation learning model for images. To approach this idea, deep metric learning (DML), serving as building a basic feature space with a pre-defined metric, has demonstrated compelling performance over the years. DML is often implemented with a carefully crafted loss function, such as the representative triplet loss, which encourages a positive sample to be by a fixed margin closer to the anchor than the negative. Despite its efficacy, the negative samples are treated uniformly, rendering the feature space less informative since different negative samples can be largely different from the anchor. In this work, we, for the first time, propose to exploit the semantic information inherent in discrete class labels as an aid for the triplet loss. Specifically, we build a bi-level negative sampling strategy,i.e., strong negative and weak negative sampling, with the guidance of an external knowledge source, from which rich class semantics can be extracted. With several fine-grained and complementary triplet losses based on this strategy, our method is enhanced with semantic awareness for image classification. In addition, to coordinate with the complicated training dynamics, we devise an ad-hoc Semantic Relation Weighting module, which consistently inspects model states and dynamically adjusts the importance of each triplet loss. It is worth noting that our method is plug-and-play, and we thus test its validity over various backbones and knowledge sources. Both qualitative and quantitative experimental results on benchmark datasets demonstrate the effectiveness of employing semantics for image classification.

Man is by nature a social animal. One important facet of human evolution is through narrative imagination, be it fictional or factual, and to tell the tale to other individuals. The factual narrative, such as news, journalism, field report, etc., is based on real-world events and often requires extensive human efforts to create. In the era of big data where video capture devices are commonly available everywhere, a massive amount of raw videos (including life-logging, dashcam or surveillance footage) are generated daily. As a result, it is rather impossible for humans to digest and analyze these video data. This paper reviews the problem of computational narrative generation where a goal-driven narrative (in the form of text with or without video) is generated from a single or multiple long videos. Importantly, the narrative generation problem makes itself distinguished from the existing literature by its focus on a comprehensive understanding of user goal, narrative structure and open-domain input. We tentatively outline a general narrative generation framework and discuss the potential research problems and challenges in this direction. Informed by the real-world impact of narrative generation, we then illustrate several practical use cases in Video Logging as a Service platform which enables users to get more out of the data through a goal-driven intelligent storytelling AI agent.

We propose Differentiable Temporal Logic (DTL), a model-agnostic framework that introduces temporal constraints to deep networks. DTL treats the outputs of a network as a truth assignment of a temporal logic formula, and computes a temporal logic loss reflecting the consistency between the output and the constraints. We propose a comprehensive set of constraints, which are implicit in data annotations, and incorporate them with deep networks via DTL. We evaluate the effectiveness of DTL on the temporal action segmentation task and observe improved performance and reduced logical errors in the output of different task models. Furthermore, we provide an extensive analysis to visualize the desirable effects of DTL.

This paper proposes a novel model for recognizing images with composite attribute-object concepts, notably for composite concepts that are unseen during model training. We aim to explore the three key properties required by the task — relation-aware, consistent, and decoupled—to learn rich and robust features for primitive concepts that compose attribute-object pairs. To this end, we propose the Blocked Message Passing Network (BMP-Net). The model consists of two modules. The concept module generates semantically meaningful features for primitive concepts, whereas the visual module extracts visual features for attributes and objects from input images. A message passing mechanism is used in the concept module to capture the relations between primitive concepts. Furthermore, to prevent the model from being biased towards seen composite concepts and reduce the entanglement between attributes and objects, we propose a blocking mechanism that equalizes the information available to the model for both seen and unseen concepts. Extensive experiments and ablation studies on two benchmarks show the efficacy of the proposed model.

We propose the Motion Capsule Autoencoder (MCAE), which addresses a key challenge in the unsupervised learning of motion representations: transformation invariance. MCAE models motion in a two-level hierarchy. In the lower level, a spatio-temporal motion signal is divided into short, local, and semantic-agnostic snippets. In the higher level, the snippets are aggregated to form full-length semantic-aware segments. For both levels, we represent motion with a set of learned transformation invariant templates and the corresponding geometric transformations by using capsule autoencoders of a novel design. This leads to a robust and efficient encoding of viewpoint changes. MCAE is evaluated on a novel Trajectory20 motion dataset and various real-world skeleton-based human action datasets. Notably, it achieves better results than baselines on Trajectory20 with considerably fewer parameters and state-of-the-art performance on the unsupervised skeleton-based action recognition task.

The computer vision community is witnessing an unprecedented rate of new tasks being proposed and addressed, thanks to the deep convolutional networks' capability to find complex mappings from X to Y. The advent of each task often accompanies the release of a large-scale annotated dataset, for supervised training of deep network. However, it is expensive and time-consuming to manually label sufficient amount of training data. Therefore, it is important to develop algorithms that can leverage off-the-shelf labeled dataset to learn useful knowledge for the target task. While previous works mostly focus on transfer learning from a single source, we study multi-source transfer across domains and tasks (MS-DTT), in a semi-supervised setting. We propose GradMix, a model-agnostic method applicable to any model trained with gradient-based learning rule, to transfer knowledge via gradient descent by weighting and mixing the gradients from all sources during training. GradMix follows a meta-learning objective, which assigns layer-wise weights to the source gradients, such that the combined gradient follows the direction that minimize the loss for a small set of samples from the target dataset. In addition, we propose to adaptively adjust the learning rate for each mini-batch based on its importance to the target task, and a pseudo-labeling method to leverage the unlabeled samples in the target domain. We conduct MS-DTT experiments on two tasks: digit recognition and action recognition, and demonstrate the advantageous performance of the proposed method against multiple baselines.

In this work, we aim to leverage prior symbolic knowledge to improve the performance of deep models. We propose a graph embedding network that projects propositional formulae (and assignments) onto a manifold via an augmented Graph Convolutional Network (GCN). To generate semantically-faithful embeddings, we develop techniques to recognize node heterogeneity, and semantic regularization that incorporate structural constraints into the embedding. Experiments show that our approach improves the performance of models trained to perform entailment checking and visual relation prediction. Interestingly, we observe a connection between the tractability of the propositional theory representation and the ease of embedding. Future exploration of this connection may elucidate the relationship between knowledge compilation and vector representation learning.