Large Multimodal Models (LMMs) extend Large Language Models (LLMs) by handling diverse inputs such as images, audio, and video, but at the cost of adding a multimodal encoding stage that increases both computational and memory overhead. This step negatively affects key Service Level Objectives (SLOs), such as time to first token (TTFT) and time per output token (TPOT). We introduce Encode-Prefill-Decode (EPD) Disaggregation, a novel framework that separates the encoding, prefill, and decode stages onto dedicated resources. Unlike current systems, which bundle encoding and prefill together, our approach decouples these steps, unlocking new opportunities and optimizations. These include a mechanism to cache multimedia tokens for efficient transfer, a novel way to parallelize the encoding load within a request, a module for optimal resource allocation for disaggregated serving, and a novel role-switching method to handle changing workload characteristics. Experimental evaluations with popular LMMs show substantial gains in memory efficiency (up to 15$\times$ lower peak memory utilization), batch sizes (up to 22$\times$ larger), 10$\times$ more images per request, and 2.2$\times$ larger KV caches. Furthermore, it leads to significant improvements in SLO attainment (up to 90–100% improvement) and TTFT (up to 71% reduction), compared to systems that do not disaggregate. The code is available at https://github.com/vbdi/epdserve.

Math Word Problem (MWP) is a crucial NLP task aimed at providing solutions for given mathematical descriptions. A notable sub-category of MWP is the Linear Programming Word Problem (LPWP), which holds significant relevance in real-world decision-making and operations research. While the recent rise of generative large language models (LLMs) has brought more advanced solutions to LPWPs, existing evaluation methodologies for this task still diverge from human judgment and face challenges in recognizing mathematically equivalent answers. In this paper, we introduce a novel evaluation metric rooted in graph edit distance, featuring benefits such as permutation invariance and more accurate program equivalence identification. Human evaluations empirically validate the superior efficacy of our proposed metric when particularly assessing LLM-based solutions for LPWP.

Federated learning is an emerging technology for training machine learning models across decentralized data sources without sharing data. Vertical federated learning, also known as feature-based federated learning, applies to scenarios where data sources have the same sample IDs but different feature sets. To ensure fairness among data owners, it is critical to objectively assess the contributions from different data sources and compensate the corresponding data owners accordingly. The Shapley value is a provably fair contribution valuation metric originating from cooperative game theory. However, its straight-forward computation requires extensively retraining a model on each potential combination of data sources, leading to prohibitively high communication and computation overheads due to multiple rounds of federated learning. To tackle this challenge, we propose a contribution valuation metric called vertical federated Shapley value (VerFedSV) based on the classic Shapley value. We show that VerFedSV not only satisfies many desirable properties of fairness but is also efficient to compute. Moreover, VerFedSV can be adapted to both synchronous and asynchronous vertical federated learning algorithms. Both theoretical analysis and extensive experimental results demonstrate the fairness, efficiency, adaptability, and effectiveness of VerFedSV.

This paper addresses the problem of ranking pre-trained models for object detection and image classification. Selecting the best pre-trained model by fine-tuning is an expensive and time-consuming task. Previous works have proposed transferability estimation based on features extracted by the pre-trained models. We argue that quantifying whether the target dataset is in-distribution (IND) or out-of-distribution (OOD) for the pre-trained model is an important factor in the transferability estimation. To this end, we propose ETran, an energy-based transferability assessment metric, which includes three scores: 1) energy score, 2) classification score, and 3) regression score. We use energy-based models to determine whether the target dataset is OOD or IND for the pre-trained model. In contrast to the prior works, ETran is applicable to a wide range of tasks including classification, regression, and object detection (classification+regression). This is the first work that proposes transferability estimation for object detection task. Our extensive experiments on four benchmarks and two tasks show that ETran outperforms previous works on object detection and classification benchmarks by an average of 21% and 12%, respectively, and achieves SOTA in transferability assessment. Code is available here1.

The simplex method, introduced by Dantzig more than half a century ago, is still to date one of the most efficient methods for solving large-scale linear programming (LP) problems. While the simplex method is known to have the finite termination property under mild assumptions, the number of iterations until optimality largely depends on the choice of initial basis. Existing strategies for selecting an advanced initial basis are mostly rule-based. These rules usually require extensive expert knowledge and empirical study to develop. Yet, many of them fail to exhibit consistent improvement, even for LP problems that arise in a single application scenario. In this paper, we propose a learning-based approach for initial basis selection. We employ graph neural networks as a building block and develop a model that attempts to capture the relationship between LP problems and their optimal bases. In addition, during the inference phase, we supplement the learning-based prediction with linear algebra tricks to ensure the validity of the generated initial basis. We validate the effectiveness of our proposed strategy by extensively testing it with state-of-the-art simplex solvers, including the open-source solver HiGHS and the commercial solver OptVerse. Through these rigorous experiments, we demonstrate that our strategy achieves substantial speedup and consistently outperforms existing rule-based methods. Furthermore, we extend the proposed approach to generating restricted master problems for column generation methods and present encouraging numerical results.

Multi-exit architectures, in which a sequence of intermediate classifiers are introduced at different depths of the feature layers, perform adaptive computation by early exiting ``easy" samples to speed up the inference. In this paper, a novel Harmonized Dense Knowledge Distillation (HDKD) training method for multi-exit architecture is designed to encourage each exit to flexibly learn from all its later exits. In particular, a general dense knowledge distillation training objective is proposed to incorporate all possible beneficial supervision information for multi-exit learning, where a harmonized weighting scheme is designed for the multi-objective optimization problem consisting of multi-exit classification loss and dense distillation loss. A bilevel optimization algorithm is introduced for alternatively updating the weights of multiple objectives and the multi-exit network parameters. Specifically, the loss weighting parameters are optimized with respect to its performance on validation set by gradient descent. Experiments on CIFAR100 and ImageNet show that the HDKD strategy harmoniously improves the performance of the state-of-the-art multi-exit neural networks. Moreover, this method does not require within architecture modifications and can be effectively combined with other previously-proposed training techniques and further boosts the performance.

Cross domain weakly supervised object detection (CDWSOD), where we can get access to instance-level annotations in the source domain while only image-level annotations are available in the target domain, adapts object detectors from label-rich to label-poor domains. It usually generates pseudo labels in the target domain and utilize them to finetune the detector pretrained in the source domain. In this paper, we propose a new pseudo-label generation-evaluation framework for CDWSOD task. In particular, an evaluator is introduced for the generated pseudo labels in the target domain and the transferring process involves two players: the detector to generate instance-level pseudo labels and the evaluator to judge the quality of pseudo labels. Only high-quality pseudo labels selected by the evaluator are utilized to finetune the detector. Experiments on three representative datasets demonstrate the effectiveness of our framework in various domains.

Muiti-exit architectures, in which a sequence of intermediate classifiers are introduced at different depths of the feature layers, perform adaptive computation by early exiting “easy” samples to speed up the inference. In this paper, we propose a new gradient deconfliction-based training technique for multi-exit architectures. In particular, the conflicting between the gradients back-propagated from different classifiers is removed by projecting the gradient from one classifier onto the normal plane of the gradient from the other classifier. Experiments on CFAR-100 and ImageNet show that the gradient deconfliction-based training strategy significantly improves the performance of the state-of-the-art multi-exit neural networks. Moreover, this method does not require within architecture modifications and can be effectively combined with other previously-proposed training techniques and further boosts the performance.