Laboratory-Scale AI: Open-Weight Models are Competitive with ChatGPT Even in Low-Resource Settings

The rapid proliferation of generative AI has raised questions about the competitiveness of lower-parameter, locally tunable, open-weight models relative to high-parameter, API-guarded, closed-weight models in terms of performance, domain adaptation, cost, and generalization. Centering under-resourced yet risk-intolerant settings in government, research, and healthcare, we see for-profit closed-weight models as incompatible with requirements for transparency, privacy, adaptability, and standards of evidence. Yet the performance penalty in using open-weight models, especially in low-data and low-resource settings, is unclear. We assess the feasibility of using smaller, open-weight models to replace GPT-4-Turbo in zero-shot, few-shot, and fine-tuned regimes, assuming access to only a single, low-cost GPU. We assess value-sensitive issues around bias, privacy, and abstention on three additional tasks relevant to those topics. We find that with relatively low effort, very low absolute monetary cost, and relatively little data for fine-tuning, small open-weight models can achieve competitive performance in domain-adapted tasks without sacrificing generality. We then run experiments considering practical issues in bias, privacy, and hallucination risk, finding that open models offer several benefits over closed models. We intend this work as a case study in understanding the opportunity cost of reproducibility and transparency over for-profit state-of-the-art zero shot performance, finding this cost to be marginal under realistic settings.

Characterizing LLM Abstention Behavior in Science QA with Context Perturbations

The correct model response in the face of uncertainty is to abstain from answering a question so as not to mislead the user. In this work, we study the ability of LLMs to abstain from answering context-dependent science questions when provided insufficient or incorrect context. We probe model sensitivity in several settings: removing gold context, replacing gold context with irrelevant context, and providing additional context beyond what is given. In experiments on four QA datasets with four LLMs, we show that performance varies greatly across models, across the type of context provided, and also by question type; in particular, many LLMs seem unable to abstain from answering boolean questions using standard QA prompts. Our analysis also highlights the unexpected impact of abstention performance on QA task accuracy. Counter-intuitively, in some settings, replacing gold context with irrelevant context or adding irrelevant context to gold context can improve abstention performance in a way that results in improvements in task performance. Our results imply that changes are needed in QA dataset design and evaluation to more effectively assess the correctness and downstream impacts of model abstention.

InfoVisDial: An Informative Visual Dialogue Dataset by Bridging Large Multimodal and Language Models

In this paper, we build a visual dialogue dataset, named InfoVisDial, which provides rich informative answers in each round even with external knowledge related to the visual content. Different from existing datasets where the answer is compact and short, InfoVisDial contains long free-form answers with rich information in each round of dialogue. For effective data collection, the key idea is to bridge the large-scale multimodal model (e.g., GIT) and the language models (e.g., GPT-3). GIT can describe the image content even with scene text, while GPT-3 can generate informative dialogue based on the image description and appropriate prompting techniques. With such automatic pipeline, we can readily generate informative visual dialogue data at scale. Then, we ask human annotators to rate the generated dialogues to filter the low-quality conversations.Human analyses show that InfoVisDial covers informative and diverse dialogue topics: $54.4\%$ of the dialogue rounds are related to image scene texts, and $36.7\%$ require external knowledge. Each round's answer is also long and open-ended: $87.3\%$ of answers are unique with an average length of $8.9$, compared with $27.37\%$ and $2.9$ in VisDial. Last, we propose a strong baseline by adapting the GIT model for the visual dialogue task and fine-tune the model on InfoVisDial. Hopefully, our work can motivate more effort on this direction.

Top-down Green-ups: Satellite Sensing and Deep Models to Predict Buffelgrass Phenology

An invasive species of grass known as "buffelgrass" contributes to severe wildfires and biodiversity loss in the Southwest United States. We tackle the problem of predicting buffelgrass "green-ups" (i.e. readiness for herbicidal treatment). To make our predictions, we explore temporal, visual and multi-modal models that combine satellite sensing and deep learning. We find that all of our neural-based approaches improve over conventional buffelgrass green-up models, and discuss how neural model deployment promises significant resource savings.

Urban Spatiotemporal Data Synthesis via Neural Disaggregation

The level of granularity of open data often conflicts the benefits it can provide. Less granular data can protect individual privacy, but to certain degrees, sabotage the promise of open data to promote transparency and assist research. Similar in the urban setting, aggregated urban data at high-level geographic units can mask out the underline particularities of city dynamics that may vary at lower areal levels. In this work, we aim to synthesize fine-grained, high resolution urban data, by breaking down aggregated urban data at coarse, low resolution geographic units. The goal is to increase the usability and realize the values as much as possible of highly aggregated urban data. To address the issue of simplicity of some traditional disaggregation methods -- 1) we experimented with numerous neural-based models that are capable of modeling intricate non-linear relationships among features. Neural methods can also leverage both spatial and temporal information concurrently. We showed that all neural methods perform better than traditional disaggregation methods. Incorporating the temporal information further enhances the results. 2) We proposed a training approach for disaggregation task, Chain-of-Training (COT), that can be incorporated into any of the training-based models. COT adds transitional disaggregation steps by incorporating intermediate geographic dimensions, which enhances the predictions at low geographic level and boosts the results at higher levels. 3) We adapted the idea of reconstruction (REC) from super-resolution domain in our disaggregation case -- after disaggregating from low to high geographic level, we then re-aggregate back to the low level from our generated high level values. Both strategies improved disaggregation results on three datasets and two cities we tested on.

Adapting to Skew: Imputing Spatiotemporal Urban Data with 3D Partial Convolutions and Biased Masking

We adapt image inpainting techniques to impute large, irregular missing regions in urban settings characterized by sparsity, variance in both space and time, and anomalous events. Missing regions in urban data can be caused by sensor or software failures, data quality issues, interference from weather events, incomplete data collection, or varying data use regulations; any missing data can render the entire dataset unusable for downstream applications. To ensure coverage and utility, we adapt computer vision techniques for image inpainting to operate on 3D histograms (2D space + 1D time) commonly used for data exchange in urban settings. Adapting these techniques to the spatiotemporal setting requires handling skew: urban data tend to follow population density patterns (small dense regions surrounded by large sparse areas); these patterns can dominate the learning process and fool the model into ignoring local or transient effects. To combat skew, we 1) train simultaneously in space and time, and 2) focus attention on dense regions by biasing the masks used for training to the skew in the data. We evaluate the core model and these two extensions using the NYC taxi data and the NYC bikeshare data, simulating different conditions for missing data. We show that the core model is effective qualitatively and quantitatively, and that biased masking during training reduces error in a variety of scenarios. We also articulate a tradeoff in varying the number of timesteps per training sample: too few timesteps and the model ignores transient events; too many timesteps and the model is slow to train with limited performance gain.

Contrastive Language-Vision AI Models Pretrained on Web-Scraped Multimodal Data Exhibit Sexual Objectification Bias

Nine language-vision AI models trained on web scrapes with the Contrastive Language-Image Pretraining (CLIP) objective are evaluated for evidence of a bias studied by psychologists: the sexual objectification of girls and women, which occurs when a person's human characteristics are disregarded and the person is treated as a body or a collection of body parts. A first experiment uses standardized images of women from the Sexual OBjectification and EMotion Database, and finds that, commensurate with prior research in psychology, human characteristics are disassociated from images of objectified women: the model's recognition of emotional state is mediated by whether the subject is fully or partially clothed. Embedding association tests (EATs) return significant effect sizes for both anger (d >.8) and sadness (d >.5). A second experiment measures the effect in a representative application: an automatic image captioner (Antarctic Captions) includes words denoting emotion less than 50% as often for images of partially clothed women than for images of fully clothed women. A third experiment finds that images of female professionals (scientists, doctors, executives) are likely to be associated with sexual descriptions relative to images of male professionals. A fourth experiment shows that a prompt of "a [age] year old girl" generates sexualized images (as determined by an NSFW classifier) up to 73% of the time for VQGAN-CLIP (age 17), and up to 40% of the time for Stable Diffusion (ages 14 and 18); the corresponding rate for boys never surpasses 9%. The evidence indicates that language-vision AI models trained on automatically collected web scrapes learn biases of sexual objectification, which propagate to downstream applications.

Data Management for Causal Algorithmic Fairness

Fairness is increasingly recognized as a critical component of machine learning systems. However, it is the underlying data on which these systems are trained that often reflects discrimination, suggesting a data management problem. In this paper, we first make a distinction between associational and causal definitions of fairness in the literature and argue that the concept of fairness requires causal reasoning. We then review existing works and identify future opportunities for applying data management techniques to causal algorithmic fairness.

Delineating Knowledge Domains in the Scientific Literature Using Visual Information

Figures are an important channel for scientific communication, used to express complex ideas, models and data in ways that words cannot. However, this visual information is mostly ignored in analyses of the scientific literature. In this paper, we demonstrate the utility of using scientific figures as markers of knowledge domains in science, which can be used for classification, recommender systems, and studies of scientific information exchange. We encode sets of images into a visual signature, then use distances between these signatures to understand how patterns of visual communication compare with patterns of jargon and citation structures. We find that figures can be as effective for differentiating communities of practice as text or citation patterns. We then consider where these metrics disagree to understand how different disciplines use visualization to express ideas. Finally, we further consider how specific figure types propagate through the literature, suggesting a new mechanism for understanding the flow of ideas apart from conventional channels of text and citations. Our ultimate aim is to better leverage these information-dense objects to improve scientific communication across disciplinary boundaries.

FairST: Equitable Spatial and Temporal Demand Prediction for New Mobility Systems

Emerging transportation modes, including car-sharing, bike-sharing, and ride-hailing, are transforming urban mobility but have been shown to reinforce socioeconomic inequities. Spatiotemporal demand prediction models for these new mobility regimes must therefore consider fairness as a first-class design requirement. We present FairST, a fairness-aware model for predicting demand for new mobility systems. Our approach utilizes 1D, 2D and 3D convolutions to integrate various urban features and learn the spatial-temporal dynamics of a mobility system, but we include fairness metrics as a form of regularization to make the predictions more equitable across demographic groups. We propose two novel spatiotemporal fairness metrics, a region-based fairness gap (RFG) and an individual-based fairness gap (IFG). Both quantify equity in a spatiotemporal context, but vary by whether demographics are labeled at the region level (RFG) or whether population distribution information is available (IFG). Experimental results on real bike share and ride share datasets demonstrate the effectiveness of the proposed model: FairST not only reduces the fairness gap by more than 80%, but can surprisingly achieve better accuracy than state-of-the-art yet fairness-oblivious methods including LSTMs, ConvLSTMs, and 3D CNN.