Instruction Tuning (IT), the process of training large language models (LLMs) using instruction-response pairs, has emerged as the predominant method for transforming base pre-trained LLMs into open-domain conversational agents. While IT has achieved notable success and widespread adoption, its limitations and shortcomings remain underexplored. In this paper, through rigorous experiments and an in-depth analysis of the changes LLMs undergo through IT, we reveal various limitations of IT. In particular, we show that (1) IT fails to enhance knowledge or skills in LLMs. LoRA fine-tuning is limited to learning response initiation and style tokens, and full-parameter fine-tuning leads to knowledge degradation. (2) Copying response patterns from IT datasets derived from knowledgeable sources leads to a decline in response quality. (3) Full-parameter fine-tuning increases hallucination by inaccurately borrowing tokens from conceptually similar instances in the IT dataset for generating responses. (4) Popular methods to improve IT do not lead to performance improvements over a simple LoRA fine-tuned model. Our findings reveal that responses generated solely from pre-trained knowledge consistently outperform responses by models that learn any form of new knowledge from IT on open-source datasets. We hope the insights and challenges revealed inspire future work.

Long-term forecasting presents unique challenges due to the time and memory complexity of handling long sequences. Existing methods, which rely on sliding windows to process long sequences, struggle to effectively capture long-term variations that are partially caught within the short window (i.e., outer-window variations). In this paper, we introduce a novel approach that overcomes this limitation by employing contrastive learning and enhanced decomposition architecture, specifically designed to focus on long-term variations. To this end, our contrastive loss incorporates global autocorrelation held in the whole time series, which facilitates the construction of positive and negative pairs in a self-supervised manner. When combined with our decomposition networks, our contrastive learning significantly improves long-term forecasting performance. Extensive experiments demonstrate that our approach outperforms 14 baseline models in multiple experiments over nine long-term benchmarks, especially in challenging scenarios that require a significantly long output for forecasting. Source code is available at https://github.com/junwoopark92/Self-Supervised-Contrastive-Forecsating.

We introduce an approach for analyzing the responses of dynamical systems to external perturbations that combines score-based generative modeling with the Fluctuation-Dissipation Theorem (FDT). The methodology enables accurate estimation of system responses, especially for systems with non-Gaussian statistics, often encountered in dynamical systems far from equilibrium. Such cases often present limitations for conventional approximate methods. We numerically validate our approach using time-series data from a stochastic partial differential equation where the score function is available analytically. Furthermore, we demonstrate the improved accuracy of our methodology over conventional methods and its potential as a versatile tool for understanding complex dynamical systems. Applications span disciplines from climate science and finance to neuroscience.

We concentrate on a novel human-centric image synthesis task, that is, given only one reference facial photograph, it is expected to generate specific individual images with diverse head positions, poses, and facial expressions in different contexts. To accomplish this goal, we argue that our generative model should be capable of the following favorable characteristics: (1) a strong visual and semantic understanding of our world and human society for basic object and human image generation. (2) generalizable identity preservation ability. (3) flexible and fine-grained head control. Recently, large pre-trained text-to-image diffusion models have shown remarkable results, serving as a powerful generative foundation. As a basis, we aim to unleash the above two capabilities of the pre-trained model. In this work, we present a new framework named CapHuman. We embrace the ``encode then learn to align" paradigm, which enables generalizable identity preservation for new individuals without cumbersome tuning at inference. CapHuman encodes identity features and then learns to align them into the latent space. Moreover, we introduce the 3D facial prior to equip our model with control over the human head in a flexible and 3D-consistent manner. Extensive qualitative and quantitative analyses demonstrate our CapHuman can produce well-identity-preserved, photo-realistic, and high-fidelity portraits with content-rich representations and various head renditions, superior to established baselines. Code and checkpoint will be released at https://github.com/VamosC/CapHuman.

Monitoring the distribution and size structure of long-living shrubs, such as Juniperus communis, can be used to estimate the long-term effects of climate change on high-mountain and high latitude ecosystems. Historical aerial very-high resolution imagery offers a retrospective tool to monitor shrub growth and distribution at high precision. Currently, deep learning models provide impressive results for detecting and delineating the contour of objects with defined shapes. However, adapting these models to detect natural objects that express complex growth patterns, such as junipers, is still a challenging task. This research presents a novel approach that leverages remotely sensed RGB imagery in conjunction with Mask R-CNN-based instance segmentation models to individually delineate Juniperus shrubs above the treeline in Sierra Nevada (Spain). In this study, we propose a new data construction design that consists in using photo interpreted (PI) and field work (FW) data to respectively develop and externally validate the model. We also propose a new shrub-tailored evaluation algorithm based on a new metric called Multiple Intersections over Ground Truth Area (MIoGTA) to assess and optimize the model shrub delineation performance. Finally, we deploy the developed model for the first time to generate a wall-to-wall map of Juniperus individuals. The experimental results demonstrate the efficiency of our dual data construction approach in overcoming the limitations associated with traditional field survey methods. They also highlight the robustness of MIoGTA metric in evaluating instance segmentation models on species with complex growth patterns showing more resilience against data annotation uncertainty. Furthermore, they show the effectiveness of employing Mask R-CNN with ResNet101-C4 backbone in delineating PI and FW shrubs, achieving an F1-score of 87,87% and 76.86%, respectively.

In the multiple unmanned aerial vehicle (UAV)- assisted downlink communication, it is challenging for UAV base stations (UAV BSs) to realize trajectory design and resource assignment in unknown environments. The cooperation and competition between UAV BSs in the communication network leads to a Markov game problem. Multi-agent reinforcement learning is a significant solution for the above decision-making. However, there are still many common issues, such as the instability of the system and low utilization of historical data, that limit its application. In this paper, a novel graph-attention multi-agent trust region (GA-MATR) reinforcement learning framework is proposed to solve the multi-UAV assisted communication problem. Graph recurrent network is introduced to process and analyze complex topology of the communication network, so as to extract useful information and patterns from observational information. The attention mechanism provides additional weighting for conveyed information, so that the critic network can accurately evaluate the value of behavior for UAV BSs. This provides more reliable feedback signals and helps the actor network update the strategy more effectively. Ablation simulations indicate that the proposed approach attains improved convergence over the baselines. UAV BSs learn the optimal communication strategies to achieve their maximum cumulative rewards. Additionally, multi-agent trust region method with monotonic convergence provides an estimated Nash equilibrium for the multi-UAV assisted communication Markov game.

Congestion pricing, while adopted by many cities to alleviate traffic congestion, raises concerns about widening socioeconomic disparities due to its disproportionate impact on low-income travelers. In this study, we address this concern by proposing a new class of congestion pricing schemes that not only minimize congestion levels but also incorporate an equity objective to reduce cost disparities among travelers with different willingness-to-pay. Our analysis builds on a congestion game model with heterogeneous traveler populations. We present four pricing schemes that account for practical considerations, such as the ability to charge differentiated tolls to various traveler populations and the option to toll all or only a subset of edges in the network. We evaluate our pricing schemes in the calibrated freeway network of the San Francisco Bay Area. We demonstrate that the proposed congestion pricing schemes improve both efficiency (in terms of reduced average travel time) and equity (the disparities of travel costs experienced by different populations) compared to the current pricing scheme. Moreover, our pricing schemes also generate a total revenue comparable to the current pricing scheme. Our results further show that pricing schemes charging differentiated prices to traveler populations with varying willingness-to-pay lead to a more equitable distribution of travel costs compared to those that charge a homogeneous price to all.

Creativity is core to being human. Generative artificial intelligence (GenAI) -- including ever more powerful large language models (LLMs) -- holds promise for humans to be more creative by offering new ideas, or less creative by anchoring on GenAI ideas. We study the causal impact of GenAI ideas on the production of a short story in an online experimental study where some writers could obtain story ideas from a GenAI platform. We find that access to GenAI ideas causes stories to be evaluated as more creative, better written, and more enjoyable, especially among less creative writers. However, GenAI-enabled stories are more similar to each other than stories by humans alone. These results point to an increase in individual creativity at the risk of losing collective novelty. This dynamic resembles a social dilemma: with GenAI, individual writers are better off, but collectively a narrower scope of novel content may be produced. Our results have implications for researchers, policy-makers and practitioners interested in bolstering creativity.

Training large language models (LLMs) with a large and diverse instruction dataset aligns the models to comprehend and follow human instructions. Recent works have shown that using a small set of high-quality instructions can outperform using large yet more noisy ones. Because instructions are unlabeled and their responses are natural text, traditional active learning schemes with the model's confidence cannot be directly applied to the selection of unlabeled instructions. In this work, we propose a novel method for instruction selection, called SelectLLM, that leverages LLMs for the selection of high-quality instructions. Our high-level idea is to use LLMs to estimate the usefulness and impactfulness of each instruction without the corresponding labels (i.e., responses), via prompting. SelectLLM involves two steps: dividing the unlabelled instructions using a clustering algorithm (e.g., CoreSet) to multiple clusters, and then prompting LLMs to choose high-quality instructions within each cluster. SelectLLM showed comparable or slightly better performance on the popular instruction benchmarks, compared to the recent state-of-the-art selection methods. All code and data are publicly available (https://github.com/minnesotanlp/select-llm).

TRopIcal DEep-sea Neutrino Telescope (TRIDENT) is a next-generation neutrino telescope to be located in the South China Sea. With a large detector volume and the use of advanced hybrid digital optical modules (hDOMs), TRIDENT aims to discover multiple astrophysical neutrino sources and probe all-flavor neutrino physics. The reconstruction resolution of primary neutrinos is on the critical path to these scientific goals. We have developed a novel reconstruction method based on graph neural network (GNN) for TRIDENT. In this paper, we present the reconstruction performance of the GNN-based approach on both track- and shower-like neutrino events in TRIDENT.