Singapore

AI inference scaling is often tuned through 1D heuristics (a fixed reasoning passes) or 2D bivariate trade-offs (e.g., performance vs.\ compute), which fail to consider cost and latency constraints. We introduce a 3D optimization framework that jointly calibrates accuracy, cost, and latency within a unified decision space, enabling constraints-aware inference scaling. Using Monte Carlo simulations across three representative scenarios and nine simulated large language models, we evaluate four optimization methods to address the 3D multi-objective optimization (MOO) problem. Framing inference scaling in MOO shapes a feasible space that 1D and 2D optimizations fail to capture, enabling environment-adaptive selection of the inference scaling~$k$. Results show that knee-point optimization achieves the best balance, while accuracy-maximization remains favorable when precision is prioritized. The framework establishes a theoretical foundation for deployment-aware inference scaling across diverse operational contexts.

AAAI 2026

3D Optimization for AI Inference Scaling: Balancing Accuracy, Cost, and Latency

artificial intelligence

optimization

inference

workshop paper

We are pleased to announce the Fortieth AAAI Conference on Artificial Intelligence (AAAI-26), which will be held in Singapore EXPO from January 20 to January 27, 2026.

The purpose of the AAAI conference series is to promote research in Artificial Intelligence (AI) and foster scientific exchange between researchers, practitioners, scientists, students, and engineers across the entirety of AI and its affiliated disciplines. AAAI-26 will feature technical paper presentations, special tracks, invited speakers, workshops, tutorials, poster sessions, senior member presentations, competitions, and exhibit programs, and a range of other activities to be announced.<br><br>

To access this event page, you need to log in with the **email address you registered with**. <br>Access credentials will be sent to your email from Underline -  subject line "Welcome to AAAI 2026". Please be sure to check your spam email folder if you do not see an email confirmation right away.

Please log in

To access this event page, you are required to register.
Please complete your registration to continue.

We recommend reading [**the registration information**](https://aaai.org/conference/aaai/aaai-26/registration/) first.

**Online Registration Form**: https://aaai.getregistered.net/conference-2026 

Registration Required

We are pleased to announce the Fortieth AAAI Conference on Artificial Intelligence (AAAI-26), which will be held in Singapore EXPO from January 20 to January 27, 2026.

Vision Language Models (VLMs) are rapidly advancing in multimodal understanding, largely due to increases in training data and model size. However, the computational demands of VLMs are also increasing, presenting challenges for their deployment in resource-constrained environments like smart televisions (TVs). VLMs are particularly useful in smart TVs for applications such as AI summarization, content question-and-answer (QnA), and user interface (UI) understanding and navigation. Previous research has focused on leveraging layer redundancy to accelerate VLM inference, but these methods often depend on extensive training, which requires significant computational resources and time. Recently, some inference-based solutions for layer skipping in Large Language Models (LLMs) have been proposed. In this paper, we demonstrate that the metrics used for layer skipping in LLMs do not always yield favorable results when applied to VLMs. We introduce an inference-only layer skipping strategy for VLMs based on attention to image tokens. Our approach increases overall throughput by up to 21% while maintaining close to baseline performance.

FlashVLM: Exploiting Layer Redundancy via Visual Attention for Efficient Vision-Language Inference

Accurate long-term traffic forecasting remains a critical challenge in intelligent transportation systems, particularly when predicting high-frequency traffic phenomena such as shock waves and congestion boundaries over extended rollout horizons. Neural operators have recently gained attention as promising tools for modeling traffic flow. While effective at learning function space mappings, they inherently produce smooth predictions that fail to reconstruct high-frequency features such as sharp density gradients which results in rapid error accumulation during multi-step rollout predictions essential for real-time traffic management. To address these fundamental limitations, we introduce a unified Diffusion-Enhanced Transformer Neural Operator (DETNO) architecture. DETNO leverages a transformer neural operator with cross-attention mechanisms, providing model expressivity and super-resolution, coupled with a diffusion-based refinement component that iteratively reconstructs high-frequency traffic details through progressive denoising. This overcomes the inherent smoothing limitations and rollout instability of standard neural operators. Through comprehensive evaluation on chaotic traffic datasets, our method demonstrates superior performance in extended rollout predictions compared to traditional and transformer-based neural operators, preserving high-frequency components and improving stability over long prediction horizons.

DETNO: A DIFFUSION-ENHANCED TRANSFORMER NEURALOPERATOR FOR LONG-TERM TRAFFIC FORECASTING

Learned database query optimizers typically optimize over a set of configurations, which limits the attainable plan space. Optimus expands the action space itself by mining novel execution plan rewrites and learns to select among these actions online. Optimus utilizes graph-based inductive matrix completion and a multilayer perceptron with the objective of minimizing latency. Crucially, the system is deployable by design: it requires no engine modifications and its rules include guard-checked compilation. On the extended JOB benchmark, Optimus yields a speedup of 1.16x over vanilla PostgreSQL solely using novel rewrites.

Optimus: Deployable Query Optimization via Novel SQL Rewrites

Although Large Language Models (LLMs) have expanded the capabilities of recommender systems, they are hindered by inherent limitations, including a propensity for factual hallucination and reliance on outdated domain knowledge. These constraints pose significant challenges in contexts requiring high-fidelity recommendations, such as consumer electronics purchases where precise, current specifications are critical. To address these issues, this paper proposes a novel framework termed Dynamic Graph-Retrieval Augmented Generation (Graph-RAG), which integrates LLMs, knowledge graphs, and RAG techniques within a multi-agent architecture. The framework dynamically deciphers complex user purchase intents and prioritizes decision-critical factors through a modular communication protocol that enables cross-agent collaboration, a `Specification Vector Index' that resolves semantic disparities between natural language queries and technical attributes, and a graph-based dynamic retrieval engine that facilitates fact-grounded reasoning. Empirical validation forms a pivotal contribution of this work, with rigorous experimental verification confirming the system's efficacy in minimizing hallucinations through structured knowledge grounding. Quantitative metrics demonstrate statistically significant improvements in recommendation accuracy, such as a +22.7% increase in precision, and reliability, while traceable decision pathways enhance operational transparency. This research delivers a foundational architecture for a possible practical recommender systems, validated through real-world deployment scenarios and test dataset, and establishes a benchmark for empirically substantiated innovation in AI-driven recommendation frameworks. By bridging theoretical innovation and practical deployment, this study marks a critical advancement in the field, offering both a new methodology and robust evidence of its real-world applicability.

Dynamic Graph-Retrieval Augmented Generation for Reliable and Explainable Consumer Electronics Recommendation

Large Language Models require substantial computational resources for inference, posing deployment challenges. While dynamic pruning offers superior efficiency over static methods through adaptive circuit selection, it exacerbates alignment degradation by retaining only input-dependent safety-critical circuit preservation across diverse inputs. As a result, addressing these heightened alignment vulnerabilities remains critical. We introduce Alignment-Aware Probe Pruning (AAPP), a dynamic structured pruning method that adaptively preserves alignment-relevant circuits during inference, building upon Probe Pruning. Experiments on LLaMA 2-7B, Qwen2.5-14B-Instruct, and Gemma-3-12B-IT show AAPP improves refusal rates by 50% at matched compute, enabling efficient yet safety-preserving LLM deployment.

Alignment-Constrained Dynamic Pruning for LLMs: Identifying and Preserving Alignment-Critical Circuits

Aligning proprietary large language models (LLMs) with internal organizational policies has become an urgent priority as organizations increasingly deploy LLMs in sensitive domains such as legal support, finance, and medical services. Beyond generic safety filters, enterprises require reliable mechanisms to detect policy violations within their regulatory and operational frameworks, where breaches can trigger legal and reputational risks. Existing content moderation frameworks, such as guardrails, remain largely confined to
the safety domain and lack the robustness to capture nuanced organizational policies. LLM-as-a-judge and fine-tuning approaches, though flexible, introduce significant latency and lack interpretability. To address these limitations, we propose a training-free and efficient method that treats policy violation detection as an out-of-distribution (OOD) detection problem. Inspired by whitening techniques, we apply a linear transformation to decorrelate the model’s hidden activations and standardize them to zero mean and unit variance, yielding
near-identity covariance matrix. In this transformed space, we use the Euclidean norm as a compliance score to detect policy violations. The method requires only the policy text and a small number of illustrative samples, which makes it lightweight and easily deployable. On a challenging policy benchmark, our approach achieves state-of-the-art results, surpassing both existing guardrails and fine-tuned reasoning models. This work provides organizations with a practical and statistically grounded framework for policy-aware oversight of
LLMs, advancing the broader goal of trustworthy, deployable AI governance. Code implementation will be released upon
publication.

Training-Free Policy Violation Detection via Activation-Space Whitening in LLMs

Large Language Models (LLMs) exhibit impressive generative capabilities but remain vulnerable to adversarial inputs, exposing potential risks such as data leakage, harmful content generation, and jailbreak attacks. Jailbreak attacks can fool LLMs and elicit harmful output even with safety alignments and guardrails. In this work, we performed a case study using AutoDAN – an automated adversarial attack generator – to stress-test open-source LLMs. We measure baseline attack success rates (ASRs) and then apply various fine-tuning defenses to mitigate these vulnerabilities. We used AutoDAN to construct adversarial datasets and evaluate the robustness of fine-tuned LLMs. Using llama-3-8b-instruct as the base model, we apply full Supervised Fine-Tuning (SFT) AutoDAN-style attacks. Our preliminary experiments show a reduction in jailbreak success rate after fine-tuning, while maintaining usefulness and coherence for benign queries. We conclude by outlining best practices for deploying adversarially resilient LLMs in production environments and future work to continue research with adversarial attack vulnerabilities and agentic workflows.

Adversarial Testing for Large Language Models: Evaluating and Enhancing Robustness with AutoDAN and Fine-Tuning Techniques

The performance of edge language models is fundamentally determined by the quality of their training data. To address the challenge of efficient data curation in resource-constrained environments, this study adapts and optimizes the Predictive Data Selection (Preselect) methodology. Our approach focuses on enhancing two core capabilities crucial for edge AI applications: ChatRAG, as the foundation for knowledge interaction, and Function Calling, as the basis for tool use.

By designing an evaluation ensemble that includes specialized models and training a FastText lightweight classifier, we can efficiently filter high-value training samples from massive datasets. Experimental results demonstrate that this strategy yields significant performance improvements, particularly in ChatRAG (+10.5%) and Function Calling (+10.0%). This research validates that an edge-optimized Preselect is an effective and viable strategy for enhancing targeted capabilities in edge models, ultimately proving that under resource constraints, curated data quality is a more critical driver of performance than mere data quantity.

Quality Over Quantity: Predictive Data Selection for Edge Language Models

Radiology worklists are typically processed first-in–first-out (FIFO) even when studies differ greatly in clinical urgency. We propose a pragmatic alternative: using calibrated probabilities of intracranial hemorrhage (ICH) to prioritize head CT exams for earlier reading. Using the public RSNA-ICH dataset, we train slice-level detectors, aggregate to exam-level, apply post-hoc calibration, and feed these scores into a transparent discrete-event simulator of the reading queue. The simulator quantifies how triage benefit reduction in median time-to-read (TTR) for ICH; scales with classifier AUC, workload (arrival rate), staffing, prevalence, and calibration. Across realistic loads, score-based prioritization yields substantial TTR reductions for ICH with minimal delay to non-ICH studies. We release a configuration-driven, reproducible pipeline that translates AI risk scores into operational metrics (minutes saved), enabling safe and data-driven evaluation before PACS/RIS deployment.

Translating Classifier Scores into Clinical Impact: Calibrated Risk and Queueing Simulation for AI-Assisted Radiology Worklist Triage

Meta-learning enables models to rapidly adapt to new tasks by leveraging prior experience, but its adaptation mechanisms remain opaque, especially regarding how past training tasks influence future predictions. We introduce TLXML (Task-Level eXplanation of Meta-Learning), a novel framework that extends influence functions to meta-learning settings, enabling task-level explanations of adaptation and inference. By reformulating influence functions for bi-level optimization, TLXML quantifies the contribution of each meta-training task to the adapted model’s behaviour. To ensure scalability, we propose a Gauss-Newton-based approximation that significantly reduces computational complexity from $O(pq^2)$ to $O(pq)$, where p and q denote model and meta parameters, respectively. Empirical results demonstrate that TLXML effectively ranks training tasks by their influence on downstream performance, offering concise and intuitive explanations aligned with user-level abstraction. This work provides a critical step toward interpretable and trustworthy meta-learning systems.

Downloads

Next from AAAI 2026

FlashVLM: Exploiting Layer Redundancy via Visual Attention for Efficient Vision-Language Inference

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES