Randomly initialized sentence encoders perform well on tasks, suggesting learning is key.
problem The role of sentence encoder architectures in language tasks.
method Random initialization and fixed architecture approach to evaluate sentence encoders.
result Priors do not leverage additional information, learning is necessary.
RL for image captions improved with a language prior.
problem Learning biases and large sample space issues in RL image captioning.
method Added a language prior to constrain the action space.
result RL with the language prior module performs better in readability and speed.
AutoElicit uses LLMs to quickly create expert priors for predictive models.
problem Creating accurate priors for predictive models is time-consuming and costly.
method AutoElicit extracts knowledge from LLMs to construct priors for predictive models.
result AutoElicit yields priors that reduce error and save labelling effort.
Improved language identification with tuplemax loss.
problem Language identification with user-specified small sets of languages.
method Replaced softmax loss with tuplemax loss.
result 2.33% error rate improvement over softmax loss.
PIPA aligns preferences without reinforcement learning, improving language model performance.
problem Aligning preferences in language models efficiently and without reinforcement learning.
method Formulates preference alignment as a Maximum Likelihood Estimation problem with prior constraints.
result PIPA algorithms achieve up to 10% performance improvement on benchmarks.
Paper tackles cold-start domain adaptation with language descriptions.
problem Cold-start domain adaptation failure with scarce target data.
method Leverages textual descriptions to learn preferences from LLM, integrates into EM algorithm.
result Framework guides source selection for weak target signals, improves as data accumulates.
New pruning method retains model expressiveness for NLP tasks.
problem Pruning large pretrained transformer models for real-world deployment.
method Mixture Gaussian Prior Pruning (MGPP) algorithm.
result MGPP outperforms existing pruning methods in high sparsity settings.
We are increasingly surrounded by artificially intelligent technology that takes decisions and executes actions on our behalf. This creates a pressing need for general means to communicate with, instruct and guide artificial agents, with human language the most compelling means for such communication. To achieve this i…
New attack reveals memorization patterns in pre-trained LLMs.
problem Determining if a data point was part of a pre-trained LLM's training set.
method Adapts MIA statistical tests to LLM's perplexity dynamics of subsequences.
result Significantly outperforms prior approaches in membership inference attacks.
A framework for flagging content with limited data.
problem Content flagging with scarce target-language data.
method Nearest-neighbor architecture using Transformer representations.
result Significant performance improvements over prior work.
Model uses LLMs to process numerical data guided by natural language descriptions.
problem Challenges in integrating prior knowledge into probabilistic models.
method Developed LLM Processes to condition numerical predictive distributions on natural language.
result Improved predictive performance and structured qualitative descriptions.
Pre-trained language model boosts RL efficiency.
problem Low sample efficiency in RL, especially in lifelong learning.
method Use a pre-trained task-independent language model for transfer learning.
result Goal-conditional RL agents become more sample efficient.
A new model tackles language generation issues by using discrete variational attention.
problem Information under-representation and posterior collapse in variational autoencoders.
method Proposes a discrete variational attention model with categorical distribution over attention mechanism.
result Enhances latent space for language generation and avoids posterior collapse.
New method shows cultural transmission improves language compositionality.
problem How cultural transmission affects language emergence and compositionality.
method Introduced cultural evolutionary dynamics into language emergence by periodically replacing agents in a population.
result Languages emerge with better compositional generalization due to implicit cultural transmission.
Paper introduces TrufLL for language model training without labeled data.
problem Training language models from scratch with reinforcement learning.
method Dynamic vocabulary truncation in RL for language generation.
result TrufLL successfully learns language generation policies from scratch.
Improved speech recognition model with better performance.
problem Speech recognition accuracy on Librispeech.
method Integrates an external language model with an internal LM correction.
result Over 14% relative improvement in performance.
Study improves keyword forecasting in earnings-call prediction markets.
problem Accurately predicting future keyword mentions in earnings calls.
method Experiments on earnings-call mention markets, varying context and market probability, introducing MCP.
result Mixture of market probability and MCP yields the best forecasts.
A new prior is proposed for learning representations of high-level concepts of the kind we manipulate with language. This prior can be combined with other priors in order to help disentangling abstract factors from each other. It is inspired by cognitive neuroscience theories of consciousness, seen as a bottleneck thro…
Unified definition of hallucinations in language models.
problem Persistent hallucinations despite mitigation efforts.
method Unified definition of hallucination as inaccurate world modeling.
result Unified framework distinguishes hallucinations from other errors.
PixL2R maps natural language to pixel-based rewards for RL, improving sample efficiency.
problem Sparse reward settings in RL limit applicability to complex problems.
method Directly maps natural language descriptions to pixel-based rewards for guiding RL.
result Language-based rewards significantly improve sample efficiency in policy learning.
We consider the scenario where the parameters of a probabilistic model are expected to vary over time. We construct a novel prior distribution that promotes sparsity and adapts the strength of correlation between parameters at successive timesteps, based on the data. We derive approximate variational inference procedur…
Recent breakthroughs in computer vision and natural language processing have spurred interest in challenging multi-modal tasks such as visual question-answering and visual dialogue. For such tasks, one successful approach is to condition image-based convolutional network computation on language via Feature-wise Linear …
LMs perform poorly in true few-shot learning without held-out examples.
problem Evaluating few-shot performance of language models without access to held-out examples.
method Evaluated two model selection criteria (cross-validation and minimum description length) for choosing LM prompts and hyperparameters in true few-shot learning.
result Selection criteria often prefer models that perform worse than random selection, suggesting overestimation of few-shot ability.
New method adapts DLMs to intrinsic data dependence without prior knowledge.
problem Understanding how unmasking schedules affect DLM generation quality.
method Adapts unmasking schedule to target data distribution's dependence structure.
result Sampling convergence guarantees improve for low-complexity distributions.
HS-MoE selects sparse experts using adaptive priors and data-adaptive gating.
problem Sparse expert selection in mixture-of-experts architectures.
method Combines horseshoe prior with input-dependent gating for data-adaptive sparsity.
result Data-adaptive sparsity in expert usage.
We introduce a novel discriminative latent variable model for bilingual lexicon induction. Our model combines the bipartite matching dictionary prior of Haghighi et al. (2008) with a representation-based approach (Artetxe et al., 2017). To train the model, we derive an efficient Viterbi EM algorithm. We provide empiric…
A single algebraic identity unifies information-theoretic variational results.
problem Deriving and generalizing classical information-theoretic variational results
method Proving a single algebraic mixed coincidence identity
result Unified derivation of classical cornerstones of information theory
Paper trains language models without memorizing user data.
problem Training production-scale language models with user privacy.
method Federated Learning with Differentially Private Federated Averaging (DP-FedAvg).
result Demonstrated training of a production NWP model with user-level differential privacy.
Study examines how decoding algorithms affect fairness in language generation models.
problem Impact of decoding algorithms on fairness in open-ended language generation.
method Systematic analysis of top-p, top-k, and temperature decoding algorithms. result Decoding algorithms significantly impact fairness across demographic groups.
Novel method diagnoses large language models' reasoning abilities.
problem Fine-grained evaluation of large language models' reasoning abilities.
method Adapting cognitive diagnosis models to LLMs, estimating mastery profiles and Q-matrix, incorporating textual information.
result Accurate parameter recovery and insights into LLMs' capabilities.
BoRA finetunes multi-task LLMs by sharing information through hierarchical priors.
problem Limited data for some tasks in multi-task LLMs.
method Bayesian hierarchical low-rank adaption.
result BoRA outperforms individual and unified model approaches.
A new watermarking method corrects bias in language models using maximal coupling.
problem Correcting bias in language model token distributions.
method Maximal coupling to balance bias correction and text quality.
result Outperforms prior techniques in preserving text quality and detectability.
APo-VAE generates text in hyperbolic space for better hierarchical representation.
problem Lack of hierarchical structure in Euclidean embeddings for natural language.
method Adversarial Poincare Variational Autoencoder (APo-VAE) in hyperbolic latent space.
result APo-VAE outperforms Euclidean VAEs in capturing latent language hierarchies.
Unified framework reduces NFEs for inverse problems.
problem High computational costs and degraded reconstruction quality in existing LDM-based inverse solvers.
method Consistency Regularised Gradient Flows for posterior sampling and prompt optimization.
result Significantly reduced computational cost with state-of-the-art performance.
A new memory-efficient sign language translation model reduces weight usage.
problem Memory constraints in real-time sign language translation.
method Variational Bayesian sequence-to-sequence network with Gaussian posterior and Indian Buffet Process prior.
result The proposed model achieves substantial weight compression without compromising performance.
Language recognition system is typically trained directly to optimize classification error on the target language labels, without using the external, or meta-information in the estimation of the model parameters. However labels are not independent of each other, there is a dependency enforced by, for example, the langu…
Adaptive querying learns user psychometrics with AI personas.
problem Learning user psychometrics within query budgets.
method Persona-induced latent variable model with AI personas and large language model response distributions.
result Persona-based posteriors deliver accurate probabilistic predictions.
Bayesian neural network achieves nearly optimal performance in Besov space.
problem Bayesian neural networks in Besov space.
method Spike-and-slab prior and shrinkage prior for posterior convergence rate.
result The posterior convergence rate is nearly minimax and adaptive to unknown smoothness.
Unified framework for convergence of discrete diffusion models without state space size dependence.
problem Fundamental limitations in existing convergence theory for discrete diffusion models, especially under singular priors and large vocabularies.
method Unified adjoint-equation-based framework that establishes dimension-free convergence guarantees in any integral probability metric (IPM).
result First dimension-free convergence bounds applicable to both masked and uniform priors, free of state space size S. We use the language of uninformative Bayesian prior choice to study the selection of appropriately simple effective models. We advocate for the prior which maximizes the mutual information between parameters and predictions, learning as much as possible from limited data. When many parameters are poorly constrained by …
New method improves RL/IL agents' adaptability to unseen environments.
problem Current RL/IL techniques struggle with generalizing to unseen environments.
method Zero-shot compositional policy learning with multi-modal fusion and attention mechanism.
result Language grounding enhances generalization across varied environments.
New methods improve integration of external LMs with AED models.
problem Improving performance of AED models by integrating external LMs.
method Comparing and proposing novel methods to estimate implicit LM from AED models.
result Proposed methods outperform previous approaches.
PMI-Masking improves MLM pretraining by masking correlated spans efficiently.
problem Uniform token masking leads to inefficient and suboptimal performance in MLMs.
method PMI-Masking uses Pointwise Mutual Information to mask n-grams with high collocation.
result PMI-Masking reaches half the training time and improves performance.
Improves cross-lingual NER by projecting entities from one language to another.
problem Limited annotated corpora for named entity recognition in many languages.
method Uses machine translation twice: first for sentences, then for entities; matches based on ortho- and phonetic similarity; identifies matches using distributional statistics.
result Improves cross-lingual NER by an average of 4.1 points on 5 diverse languages.
Paper proposes using logic networks to inject prior knowledge for better reinforcement learning.
problem Improving reinforcement learning agents with prior knowledge of object and event semantics.
method Integrates first-order logic grounded in deep neural networks as prior knowledge into reinforcement learning algorithms.
result Demonstrates that combining symbolic and image layers in a single decision module improves learning efficiency.
Dynamic meta-learning improves multi-agent communication with natural language.
problem Static population-based approaches struggle with adapting to human partners using natural language.
method Dynamic population-based meta-learning approach.
result Agents outperform all prior work in communicating with seen partners and humans.
Statsformer validates and adapts LLM-derived semantic priors for improved supervised learning.
problem Unreliable semantic priors from LLMs can degrade supervised learning performance.
method Adapts LLM-derived feature scores into a family of learner-specific prior-injection mechanisms, calibrating their influence using out-of-fold validation.
result Improves prediction performance by adaptively downweighting unreliable LLM priors, ensuring a guardrailed statistical learning system.
Many methods have been used to recognize author personality traits from text, typically combining linguistic feature engineering with shallow learning models, e.g. linear regression or Support Vector Machines. This work uses deep-learning-based models and atomic features of text, the characters, to build hierarchical, …