Rethinking Token Reweighting for SFT: Suppress, Reverse, and Extrapolate Learned Features

Existing token-reweighting methods cannot reverse harmful SFT features; SCALE uses frozen SFT deltas with entropy-guided gates to suppress, reverse, or extrapolate them, improving math and code results.

Cunchun Li, Haonan He, Yifan Gao, Minglei Li and 3 more

Published Sep 27, 2026 · 0 citations · ▲ 11 on Hugging Face

LITE: A Lightweight Lazy Sampler for Efficient SGD

Amir Daghestani, Mikael Johansson

Sydney Poster Session 6, Thu, Dec 10, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026

Exact power indices for plurality-voting ensembles

Ilie Sarpe, Theofanis Georgakopoulos, Aristides Gionis

Sydney Poster Session 2, Tue, Dec 8, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026

Speeding up Log-Sum-Exp: Kernel Fusion at the Memory Wall, Integer Arithmetic at the Compute Wall

Lingyun Yao, Martin Andraud, Niki Loppi, Andrea Pilzer and 3 more

Paris Poster Session 4, Thu, Dec 10, 5:30 PM–7:30 PM, Paris Poster Hall · Published 2026

Differentiable Systematic Resampling for Variational Sequential Monte Carlo

Fredrik Cumlin, Saikat Chatterjee

Atlanta Poster Session 3, Thu, Dec 10, 10:00 AM–1:00 PM, Hall C1 · Published 2026

A Matter of Interest: Understanding Interestingness Judgments of Math Problems in Humans and Language Models

Shubhra Mishra, Yuka Machino, Gabriel Poesia, Albert Q. Jiang and 8 more

Sydney Poster Session 3, Wed, Dec 9, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026

Robust Flow Matching under Target Corruption and Label Noise

Mert Can Kurucu, Erik Englesson, Hossein Azizpour

Paris Poster Session 4, Thu, Dec 10, 5:30 PM–7:30 PM, Paris Poster Hall · Published 2026

The Adversarial Gait: Detecting Visual Adversarial Attacks against Vision-Language Models via Self-Targeted Gradient Characterization

Mauricio Byrd Victorica, Ezzeldin Shereen, György Dán

Sydney Poster Session 3, Wed, Dec 9, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026

State of Thought Enables Endogenous Reasoning

SoT enables endogenous LLM reasoning via internal dynamics-geometric states and a lightweight controller, improving accuracy by up to 2.51x while reducing tokens by 62.6% and latency by 44.6% versus external reasoning methods.

Zhiren Gong, Yikun Hou, Zihao Zeng, Ming Xiao and 2 more

Sydney Poster Session 3, Wed, Dec 9, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026

PrismFlow: Residual Dynamics for Flow Matching in Time-Series Generation

PrismFlow uses Koopman-inspired dynamical experts with a confidence-aware winner-take-all objective to learn residual flow corrections that recover fine-grained temporal dynamics and mitigate spectral contraction in flow matching.

ZHANG JUNRU, Lang Feng, Jinbo Wang, Xu Guo and 5 more

Sydney Poster Session 3, Wed, Dec 9, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026

MindAlign: Bridging EEG, Vision, and Language for Zero-Shot Visual Decoding

MindAlign aligns EEG, vision, and language via tri-modal contrastive learning to achieve 54.1% zero-shot visual decoding accuracy on Things-EEG2.

Zexuan Chen, Sichao Liu, Runhao Lu, Huichao Qi and 3 more

Sydney Poster Session 5, Thu, Dec 10, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026

Dynamic k-center clustering with lifetimes

A dynamic k-center model with known lifetimes achieves deterministic (2+ε)-approximation with amortized updates and linear memory, plus a (6+ε)-approximation with worst-case updates and sublinear memory.

Simone Moretti, Paolo Pellizzoni, Andrea Pietracaprina, Geppino Pucci

Atlanta Poster Session 2, Wed, Dec 9, 4:30 PM–7:30 PM, Hall C1 · Published 2026

Deep Probabilistic Supervision for Image Classification

Deep Probabilistic Supervision constructs sample-specific target distributions via statistical inference on model predictions, improving accuracy, calibration, and label-noise robustness without hard targets.

Anton Adelöw, Matteo Gamba, Atsuto Maki

Sydney Poster Session 1, Tue, Dec 8, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026

Density-Ratio Losses for Post-Hoc Learning to Defer

Post-hoc learning to defer is cast as density-ratio estimation between ideal distributions, yielding adjustable deferral rules that recover Chow's rule and outperform baselines.

Alexander Soen, Ragnar Thobaben, Joakim Jaldén, Richard Nock

Sydney Poster Session 1, Tue, Dec 8, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026

Sparsely Supervised Diffusion

Sparsely supervised diffusion masks up to 98% of training pixels to fix spatial inconsistency, improve FID, reduce memorization, and stabilize small-dataset training.

Wenshuai Zhao, Zhiyuan Li, Yi Zhao, Mohammad Vali and 4 more

Sydney Poster Session 4, Wed, Dec 9, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026

Graph Cascades: Contagion-Based Mesoscopic Rewiring for Structure-Aware Graph Machine Learning

Graph Cascades uses contagion diffusion to build auxiliary edges in linear time, boosting GNN and graph transformer accuracy on heterophilic and high-degree graphs while failing on regular low-degree graphs.

Meher Chaitanya Pindiprolu, My Le, Luana Ruiz

Sydney Poster Session 4, Wed, Dec 9, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026