Parameter-efficient fine-tuning (PEFT) methods are increasingly used with pre-trained language models (PLMs) for continual learning (CL). These methods typically involve training a PEFT module for each new task and employing similarity-based selection to route modules during inference. However, they face two major limitations: 1) interference during module training with already learned modules and 2) suboptimal routing when composing modules. In this paper, we present L2R, a method that isolates the training of new PEFT modules to ensure their task specialization. L2R then learns to compose the learned modules by training a network of routers that leverages a small memory containing examples of previously seen tasks. We evaluate our method in two CL setups using various benchmarks. Our results demonstrate that L2R provides an effective composition of PEFT modules, leading to improved generalization and performance compared to other methods.

This study presents a novel evaluation framework for the Vision-Language Navigation (VLN) task. It aims to diagnose current models for various instruction categories at a finer-grained level. The framework is structured around the context-free grammar (CFG) of the task. The CFG serves as the basis for the problem decomposition and the core premise of the instruction categories design. We propose a semi-automatic method for CFG construction with the help of Large-Language Models (LLMs). Then, we induct and generate data spanning five principal instruction categories (i.e. direction change, landmark recognition, region recognition, vertical movement, and numerical comprehension). Our analysis of different models reveals notable performance discrepancies and recurrent issues. The stagnation of numerical comprehension, heavy selective biases over directional concepts, and other interesting findings contribute to the development of future language-guided navigation systems. The project is now available at https://zehao-wang.github.io/navnuances.

News Image Captioning aims to create captions from news articles and images, emphasizing the connection between textual context and visual elements. Recognizing the significance of human faces in news images and the face-name co-occurrence pattern in existing datasets, we propose a face-naming module for learning better name embeddings. Apart from names, which can be directly linked to an image area (faces), news image captions mostly contain context information that can only be found in the article. We design a retrieval strategy using CLIP to retrieve sentences that are semantically close to the image, mimicking human thought process of linking articles to images. Furthermore, to tackle the problem of the imbalanced proportion of article context and image context in captions, we introduce a simple yet effective method Contrasting with Language Model backbone (CoLaM) to the training pipeline. We conduct extensive experiments to demonstrate the efficacy of our framework. We out-perform the previous state-of-the-art (without external data) by 7.97/5.80 CIDEr scores on GoodNews/NYTimes800k. Our code is available at https://github.com/tingyu215/VACNIC.

In this paper, we aim to develop a self-supervised grounding of Covid-related medical text based on the actual spatial relationships between the referred anatomical concepts. More specifically, we learn to project sentences into a physical space defined by a three-dimensional anatomical atlas, allowing for a visual approach to navigating Covid-related literature. We design a straightforward and empirically effective training objective to reduce the curated data dependency issue. We use BERT as the main building block of our model and perform a quantitative analysis that demonstrates that the model learns a context-aware mapping. We illustrate two potential use-cases for our approach, one in interactive, 3D data exploration, and the other in document retrieval. To accelerate research in this direction, we make public all trained models, codebase and the developed tools, which can be accessed at https://github.com/gorjanradevski/macchina/.

In this paper, we develop a method for grounding medical text into a physically meaningful and interpretable space corresponding to a human atlas. We build on text embedding architectures such as Bert and introduce a loss function that allows us to reason about the semantic and spatial relatedness of medical texts by learning a projection of the embedding into a 3D space representing the human body. We quantitatively and qualitatively demonstrate that our proposed method learns a context sensitive and spatially aware mapping, in both the inter-organ and intra-organ sense, using a large scale medical text dataset from the “Large-scale online biomedical semantic indexing” track of the 2020 BioASQ challenge. We extend our approach to a self-supervised setting, and find it to be competitive with a classification based method, and a fully supervised variant of approach.

We address the problem of multimodal spatial understanding by decoding a set of language-expressed spatial relations to a set of 2D spatial arrangements in a multi-object and multi-relationship setting. We frame the task as arranging a scene of clip-arts given a textual description. We propose a simple and effective model architecture Spatial-Reasoning Bert (SR-Bert), trained to decode text to 2D spatial arrangements in a non-autoregressive manner. SR-Bert can decode both explicit and implicit language to 2D spatial arrangements, generalizes to out-of-sample data to a reasonable extent and can generate complete abstract scenes if paired with a clip-arts predictor. Finally, we qualitatively evaluate our method with a user study, validating that our generated spatial arrangements align with human expectation.

We investigate animal recognition models learned from wildlife video documentaries by using the weak supervision of the textual subtitles. This is a particularly challenging setting, since i) the animals occur in their natural habitat and are often largely occluded and ii) subtitles are to a large degree complementary to the visual content, providing a very weak supervisory signal. This is in contrast to most work on integrated vision and language in the literature, where textual descriptions are tightly linked to the image content, and often generated in a curated fashion for the task at hand. In particular, we investigate different image representations and models, including a support vector machine on top of activations of a pretrained convolutional neural network, as well as a Naive Bayes framework on a ‘bag-of-activations’ image representation, where each element of the bag is considered separately. This representation allows key components in the image to be isolated, in spite of largely varying backgrounds and image clutter, without an object detection or image segmentation step. The methods are evaluated based on how well they transfer to unseen camera-trap images captured across diverse topographical regions under different environmental conditions and illumination settings, involving a large domain shift.