Machine Learning Research Highlights 🤖

Objective

Diffusion Language Models (DLMs) incur high inference cost due to iterative denoising, motivating efficient pruning

Method

Existing pruning heuristics largely inherited from autoregressive (AR) LLMs, typically preserve attention sink tokens because AR sinks serve as stable global anchors

Results

We show that this assumption does not hold for DLMs: the attention-sink position exhibits substantially higher variance over the full generation trajectory (measured by how the dominant sink locations shift across timesteps), indicating that sinks are often transient and less structurally essential than in AR models

Significance

Based on this observation, we propose Sink-Aware Pruning, which automatically identifies and prunes unstable sinks in DLMs (prior studies usually keep sinks for AR LLMs)

ArXiv Link

Objective

HIPE-2026 is a CLEF evaluation lab dedicated to person-place relation extraction from noisy, multilingual historical texts

Method

Building on the HIPE-2020 and HIPE-2022 campaigns, it extends the series toward semantic relation extraction by targeting the task of identifying person--place associations in multiple languages and time periods

Results

Systems are asked to classify relations of two types - at ("Has the person ever been at this place?") and isAt ("Is the person located at this place around publication time?") - requiring reasoning over temporal and geographical cues

Significance

The lab introduces a three-fold evaluation profile that jointly assesses accuracy, computational efficiency, and domain generalization

ArXiv Link

Objective

Reward modeling is a core component of modern alignment pipelines including RLHF and RLAIF, underpinning policy optimization methods including PPO and TRPO

Method

However, training reliable reward models relies heavily on human-labeled preference data, which is costly and limited, motivating the use of data augmentation

Results

Existing augmentation approaches typically operate at the representation or semantic level and remain agnostic to the reward model's estimation difficulty

Significance

In this paper, we propose MARS, an adaptive, margin-aware augmentation and sampling strategy that explicitly targets ambiguous and failure modes of the reward model

ArXiv Link

Objective

We propose a two-stage "Mine and Refine" contrastive training framework for semantic text embeddings to enhance multi-category e-commerce search retrieval

Method

Large scale e-commerce search demands embeddings that generalize to long tail, noisy queries while adhering to scalable supervision compatible with product and policy constraints

Results

A practical challenge is that relevance is often graded: users accept substitutes or complements beyond exact matches, and production systems benefit from clear separation of similarity scores across these relevance strata for stable hybrid blending and thresholding

Significance

To obtain scalable policy consistent supervision, we fine-tune a lightweight LLM on human annotations under a three-level relevance guideline and further reduce residual noise via engagement driven auditing

ArXiv Link

Objective

As humans increasingly rely on multiround conversational AI for high stakes decisions, principled frameworks are needed to ensure such interactions reliably improve decision quality

Method

We adopt a human centric view governed by two principles: counterfactual harm, ensuring the AI does not undermine human strengths, and complementarity, ensuring it adds value where the human is prone to err

Results

We formalize these concepts via user defined rules, allowing users to specify exactly what harm and complementarity mean for their specific task

Significance

We then introduce an online, distribution free algorithm with finite sample guarantees that enforces the user-specified constraints over the collaboration dynamics

ArXiv Link