Kolmogorov-Arnold Networks (KANs) offer a promising path toward interpretable machine learning: their learnable activations can be studied individually, while collectively fitting complex data accurately. In practice, however, trained activations often lack symbolic fidelity, learning pathological decompositions with no meaningful correspondence to interpretable forms. We propose Softly Symbolified Kolmogorov-Arnold Networks (S2KAN), which integrate symbolic primitives directly into training. Each activation draws from a dictionary of symbolic and dense terms, with learnable gates that sparsify the representation. Crucially, this sparsification is differentiable, enabling end-to-end optimization, and is guided by a principled Minimum Description Length objective. When symbolic terms suffice, S2KAN discovers interpretable forms; when they do not, it gracefully degrades to dense splines. We demonstrate competitive or superior accuracy with substantially smaller models across symbolic benchmarks, dynamical systems forecasting, and real-world prediction tasks, and observe evidence of emergent self-sparsification even without regularization pressure.

Deep generative models (DGMs) for graphs achieve impressively high expressive power thanks to very efficient and scalable neural networks. However, these networks contain non-linearities that prevent analytical computation of many standard probabilistic inference queries, i.e., these DGMs are considered \emph{intractable}. While recently proposed Probabilistic Graph Circuits (PGCs) address this issue by enabling \emph{tractable} probabilistic inference, they operate on dense graph representations with $\mathcal{O}(n^2)$ complexity for graphs with $n$ nodes and \emph{$m$ edges}. To address this scalability issue, we introduce Sparse PGCs, a new class of tractable generative models that operate directly on sparse graph representation, reducing the complexity to $\mathcal{O}(n + m)$, which is particularly beneficial for $m \ll n^2$. In the context of de novo drug design, we empirically demonstrate that SPGCs retain exact inference capabilities, improve memory efficiency and inference speed, and match the performance of intractable DGMs in key metrics.

State-of-the-art extractive question-answering models achieve superhuman performances on the SQuAD benchmark. Yet, they are unreasonably heavy and need expensive GPU computing to answer questions in a reasonable time. Thus, they cannot be used in the open-domain question-answering paradigm for real-world queries on hundreds of thousands of documents. In this paper, we explore the possibility of transferring the natural language understanding of language models into dense vectors representing questions and answer candidates to make question-answering compatible with a simple nearest neighbor search task. This new model, which we call EfficientQA, takes advantage of the pair of sequences kind of input of BERT-based models to build meaningful, dense representations of candidate answers. These latter are extracted from the context in a question-agnostic fashion. Our model achieves state-of-the-art results in Phrase-Indexed Question Answering (PIQA), beating the previous state-of-art by 1.3 points in exact-match and 1.4 points in f1-score. These results show that dense vectors can embed rich semantic representations of sequences, although these were built from language models not originally trained for the use case. Thus, to build more resource-efficient NLP systems in the future, training language models better adapted to build dense representations of phrases is one of the possibilities.

Generative models have recently gained attention in recommendation systems by directly predicting item identifiers from user interaction sequences. However, existing methods suffer from significant information loss due to the separation of stages such as quantization and sequence modeling, hindering their ability to achieve the modeling precision and accuracy of sequential dense retrieval techniques. Integrating generative and dense retrieval methods remains a critical challenge. To address this, we introduce the Cascaded Organized Bi-Represented generAtive retrieval (COBRA) framework, which innovatively integrates sparse semantic IDs and dense vectors through a cascading process. Our method alternates between generating these representations by first generating sparse IDs, which serve as conditions to aid in the generation of dense vectors. End-to-end training enables dynamic refinement of dense representations, capturing both semantic insights and collaborative signals from user-item interactions. During inference, COBRA employs a coarse-to-fine strategy, starting with sparse ID generation and refining them into dense vectors via the generative model. We further propose BeamFusion, an innovative approach combining beam search with nearest neighbor scores to enhance inference flexibility and recommendation diversity. Extensive experiments on public datasets and offline tests validate our method's robustness. Online A/B tests on a real-world advertising platform with over 200 million daily users demonstrate substantial improvements in key metrics, highlighting COBRA's practical advantages.

Masked auto-encoder pre-training has emerged as a prevalent technique for initializing and enhancing dense retrieval systems. It generally utilizes additional Transformer decoder blocks to provide sustainable supervision signals and compress contextual information into dense representations. However, the underlying reasons for the effectiveness of such a pre-training technique remain unclear. The usage of additional Transformer-based decoders also incurs significant computational costs. In this study, we aim to shed light on this issue by revealing that masked auto-encoder (MAE) pre-training with enhanced decoding significantly improves the term coverage of input tokens in dense representations, compared to vanilla BERT checkpoints. Building upon this observation, we propose a modification to the traditional MAE by replacing the decoder of a masked auto-encoder with a completely simplified Bag-of-Word prediction task. This modification enables the efficient compression of lexical signals into dense representations through unsupervised pre-training. Remarkably, our proposed method achieves state-of-the-art retrieval performance on several large-scale retrieval benchmarks without requiring any additional parameters, which provides a 67% training speed-up compared to standard masked auto-encoder pre-training with enhanced decoding.

Data-driven weather prediction (DDWP) models are increasingly becoming popular for weather forecasting. However, while operational weather forecasts predict a wide variety of weather variables, DDWPs currently forecast a specific set of key prognostic variables. Non-prognostic ("diagnostic") variables are sometimes modeled separately as dependent variables of the prognostic variables (c.f. FourCastNet), or by including the diagnostic variable as a target in the DDWP. However, the cost of training and deploying bespoke models for each diagnostic variable can increase dramatically with more diagnostic variables, and limit the operational use of such models. Likewise, retraining an entire DDWP each time a new diagnostic variable is added is also cost-prohibitive. We present an two-stage approach that allows new diagnostic variables to be added to an end-to-end DDWP model without the expensive retraining. In the first stage, we train an autoencoder that learns to embed prognostic variables into a latent space. In the second stage, the autoencoder is frozen and "downstream" models are trained to predict diagnostic variables using only the latent representations of prognostic variables as input. Our experiments indicate that models trained using the two-stage approach offer accuracy comparable to training bespoke models, while leading to significant reduction in resource utilization during training and inference. This approach allows for new "downstream" models to be developed as needed, without affecting existing models and thus reducing the friction in operationalizing new models.

Injecting textual information into knowledge graph (KG) entity representations has been a worthwhile expedition in terms of improving performance in KG oriented tasks within the NLP community. External knowledge often adopted to enhance KG embeddings ranges from semantically rich lexical dependency parsed features to a set of relevant key words to entire text descriptions supplied from an external corpus such as wikipedia and many more. Despite the gains this innovation (Text-enhanced KG embeddings) has made, the proposal in this work suggests that it can be improved even further. Instead of using a single text description (which would not sufficiently represent an entity because of the inherent lexical ambiguity of text), we propose a multi-task framework that jointly selects a set of text descriptions relevant to KG entities as well as align or augment KG embeddings with text descriptions. Different from prior work that plugs formal entity descriptions declared in knowledge bases, this framework leverages a retriever model to selectively identify richer or highly relevant text descriptions to use in augmenting entities. Furthermore, the framework treats the number of descriptions to use in augmentation process as a parameter, which allows the flexibility of enumerating across several numbers before identifying an appropriate number. Experiment results for Link Prediction demonstrate a 5.5% and 3.5% percentage increase in the Mean Reciprocal Rank (MRR) and Hits@10 scores respectively, in comparison to text-enhanced knowledge graph augmentation methods using traditional CNNs.

Electronic Health Records (EHRs) aggregate diverse information at the patient level, holding a trajectory representative of the evolution of the patient health status throughout time. Although this information provides context and can be leveraged by physicians to monitor patient health and make more accurate prognoses/diagnoses, patient records can contain information from very long time spans, which combined with the rapid generation rate of medical data makes clinical decision making more complex. Patient trajectory modelling can assist by exploring existing information in a scalable manner, and can contribute in augmenting health care quality by fostering preventive medicine practices. We propose a solution to model patient trajectories that combines different types of information and considers the temporal aspect of clinical data. This solution leverages two different architectures: one supporting flexible sets of input features, to convert patient admissions into dense representations; and a second exploring extracted admission representations in a recurrent-based architecture, where patient trajectories are processed in sub-sequences using a sliding window mechanism. The developed solution was evaluated on two different clinical outcomes, unexpected patient readmission and disease progression, using the publicly available MIMIC-III clinical database. The results obtained demonstrate the potential of the first architecture to model readmission and diagnoses prediction using single patient admissions. While information from clinical text did not show the discriminative power observed in other existing works, this may be explained by the need to fine-tune the clinicalBERT model. Finally, we demonstrate the potential of the sequence-based architecture using a sliding window mechanism to represent the input data, attaining comparable performances to other existing solutions.