Researchers propose ALER-TI, a retrieval-augmented framework for time series imputation that addresses limitations of architectures relying solely on local temporal context. The core contribution is Latent Embedding Alignment (LEA), which mitigates representation mismatch between corrupted query sequences and complete historical candidates by applying post-hoc masking in latent space, enabling pre-computation and caching of historical embeddings. The framework is model-agnostic and demonstrated to improve strong baseline models across six real-world datasets under varying missing rates.
A new arXiv preprint introduces CADE (Contrastive Alignment with Direct Embedding), a framework for time-series question answering (TSQA) that bypasses the tokenization bottleneck of standard LLMs by mapping each timestep directly into the LLM embedding space via a point-wise linear encoder and MLP projector. The approach also introduces a one-directional supervised contrastive loss to align time-series embeddings with frozen class-name text anchors, bridging the semantic gap between numerical and language representations. Evaluated on the Time-MQA benchmark across six TSQA tasks, CADE outperforms both open-source and proprietary LLM baselines. The work addresses a concrete limitation of patch-based encoders — fixed granularity and poor cross-dataset transfer — with a cleaner architectural alternative.
TiRex-2 is a recurrent xLSTM-based time series foundation model that extends the univariate TiRex to multivariate forecasting with past and future covariates, while supporting streaming inference at constant per-patch cost. The model uses a bidirectional time mixer and asymmetric grouped-attention variate mixer to handle future-known covariates without violating causality over target variables. A synthetic coupling pipeline enables scalable multivariate pretraining from univariate corpora. TiRex-2 claims state-of-the-art zero-shot performance on GIFT-Eval and fev-bench benchmarks with 38.4M–82.5M parameters depending on mode.
TEVI is a framework that uses sparse autoencoders to disentangle CLIP image embeddings and a learned masking module to selectively reconstruct embeddings conditioned on a given caption, addressing the information imbalance between images and their captions. The approach improves image-text retrieval on both coarse-grained benchmarks (MS COCO, Flickr) and fine-grained long-caption benchmarks (IIW, DOCCI), with larger gains on richer captions. The work also shows improved robustness on the RoCOCO benchmark.
GenAIR is a framework that uses LLMs to infer 'archetype' profiles of items' ideal target audiences, generating richer item embeddings for sequential recommendation systems. A behavioral calibration objective aligns these semantic embeddings with actual user interaction patterns, closing the gap between language-space representations and real-world behavior. Experiments on three datasets show consistent improvements over state-of-the-art baselines across multiple sequential recommendation models.
A new arXiv preprint introduces Latent World Recovery (LWR), a framework for multimodal learning when some modalities are unavailable at training or inference time. LWR aligns modality-specific embeddings in a shared latent space and fuses only available modalities, avoiding explicit reconstruction of missing ones. The approach is evaluated on incomplete multi-omics benchmarks for cancer phenotype classification and survival prediction, demonstrating robustness under partial observation.
DREAM is a new method for training dense retrieval embedding models using the autoregressive next-token prediction objective of a frozen LLM, bypassing the need for labeled positive/negative document pairs required by contrastive training. The approach injects retriever-generated query-document similarity scores into selected attention heads of the LLM, allowing prediction loss gradients to flow back to the retriever. Evaluated on BEIR and RTEB benchmarks with 0.5B–3B parameter backbones, DREAM consistently outperforms contrastive baselines across model scales.
REAL is a new framework that represents LLM conversational memory as a temporal, confidence-aware directed property graph, where atomic facts carry validity intervals, confidence scores, and exploration intent labels. It addresses three limitations of prior memory systems: flat text structures, destructive overwrites of evolving facts, and passive retrieval. The system uses non-destructive temporal updates, semantic evaluator-guided hybrid beam search, and counterfactual inference to repair incomplete retrieval states. Experiments show a 22.72% average improvement over flat-text, graph-based, and existing memory baselines.
Researchers study whether instruction-following audio language models (ALMs) use explicit acoustic cues in a grounded way when raw audio is already available. They derive six interpretable acoustic concept tokens from the eGeMAPS feature set and append them to text prompts, testing on FAU-Aibo and IEMOCAP benchmarks. Aligned tokens improve unweighted average recall while shuffled or corrupted tokens degrade performance, but models don't fully collapse under perturbation, indicating partial anchoring to the audio signal. The work offers a practical probing method for interpretability and robustness in affective computing with ALMs.