Search papers, labs, and topics across Lattice.
100 papers published across 3 labs.
System-aware optimizations in KV cache design could dramatically reduce the memory footprint and costs of serving large language models.
Switching between autoregressive and diffusion modes allows Nemotron-Labs-Diffusion to achieve unprecedented throughput and efficiency in language modeling.
Significant variations in stereotypical behavior across Spanish-speaking countries reveal the limitations of English-centric stereotype datasets in AI.
Allocating budget between resampling and rerouting can dramatically enhance response quality in large language models, especially when verifier accuracy is considered.
ConOrd redefines ordinal regression by seamlessly integrating global structure into contrastive learning, outperforming existing methods across multiple benchmarks.
Allocating budget between resampling and rerouting can dramatically enhance response quality in large language models, especially when verifier accuracy is considered.
ConOrd redefines ordinal regression by seamlessly integrating global structure into contrastive learning, outperforming existing methods across multiple benchmarks.
Hierarchical structures in documents can boost analysis accuracy by preserving context, leading to more effective filtering and question answering.
Existing multilingual encoders can mislead researchers by fragmenting minority languages, but a new corpus and method reveal their true potential.
Achieving structured pruning that rivals unstructured methods in accuracy while significantly accelerating inference speed could redefine efficiency benchmarks for large language models.
Despite reducing persona collapse by 80%, LLMs still struggle to match human adaptability in advice-giving, with users favoring the default persona even in challenging situations.
LLMs can generate diverse resident personas that produce executable smart home interaction schedules, eliminating the need for intrusive real-world data collection.
Personalized prompts in robot therapy sessions can boost engagement and mitigate cognitive fatigue in dementia care.
Surpassing larger models, this agent achieves 91.4% retrieval accuracy in long-horizon multimodal dialogues by leveraging episodic memory for efficient context management.
EDO achieves up to 78% reduction in cross-entropy, revolutionizing how we model human subjectivity in NLP tasks.
Current large language models are overconfident, but a new calibration method for eigenvalues could significantly enhance their reliability in real-world applications.
Gaze-only models can be significantly improved by injecting lightweight language signals, achieving up to a 2.9 percentage point gain in reading comprehension prediction accuracy.
Achieving state-of-the-art parsing performance with 99.94% fewer rule-scoring parameters, Hol-PCFG transforms how we interpret grammar rules in unsupervised parsing.
System-aware optimizations in KV cache design could dramatically reduce the memory footprint and costs of serving large language models.
Usage of the AI learning assistant Syntea varies significantly across demographics, revealing critical insights into how different student groups interact with educational technology.
WebSwarm's innovative recursive delegation allows agents to not only search but also adaptively collaborate, leading to superior performance in complex web search tasks.
AMALIA's performance reveals that high agreement with human coders does not guarantee valid measurement of complex constructs, highlighting critical flaws in national language model evaluations.
Models can misrepresent unanswerable questions, but a new calibrated policy allows for precise control over when they should answer.
Communication style can dramatically shift triage outcomes, highlighting the risks of deploying chatbots trained on idealized patient interactions.
HCC-STAR not only surpasses leading models in treatment accuracy but also offers a significant survival advantage, highlighting the potential of AI in precision oncology.
Sum-abs reduction in SHAP-weighted fusion not only preserves attribution mass but also enhances multimodal emotion recognition, nearly matching the performance of traditional early fusion methods.
Uncovering unified psychological structures, JAM achieves superior personality recognition without the constraints of predefined taxonomies, revolutionizing how we infer psychological profiles from text.
Shifting the error landscape in compliance management, this pipeline reveals that a single misidentified asset can lead to irrelevant vulnerabilities, making risk assessment more visible and manageable.
Distilling a reasoning model into a compact on-device version recovers significant summary quality while drastically reducing processing time from 39 seconds to just 0.8 seconds per article.
Psychological competence in AI evaluation could redefine how we assess the impact of AI on human cognition and decision-making.
PolyUQuest achieves superior answer correctness and verifiability by integrating structural web data into its retrieval process, outperforming traditional RAG systems.
LLMs fall short of clinician performance in psychiatric evaluations, trailing by over 37 percentage points in objective competence.
Compact and informative schemas generated by ASMR can revolutionize the way ship maintenance reports are authored, leading to more actionable insights.
TokenWall slashes the attack success rate to 12.5% while ensuring a 97.4% pass rate for benign interactions, all with just 0.69 seconds of added latency.
MASTE achieves zero-shot Aspect Sentiment Triplet Extraction with a multi-agent approach that outperforms traditional LLM methods, even without labeled data.
SkelGen4D achieves high-quality text-driven mesh animation without the burden of extensive skeleton annotations, outperforming fully supervised models.
Log-Insight transforms incident diagnosis from a manual, time-consuming process into an efficient automated workflow, achieving over 90% accuracy in identifying root causes within a minute.
This system transforms how we interact with historical records, enabling complex queries that weave together expert knowledge and document retrieval in real-time.
Lexical and structural hashing methods can match near-duplicate documents effectively, but semantic-sensitive approaches excel in preserving similarity under content rewriting, albeit at a higher computational cost.
Tenant responses to sustainability communications are more aligned with housing providers' posts than random interactions, highlighting the power of organizational messaging in shaping public discourse.
COALA outperforms existing methods in contextual biasing for ASR, achieving superior performance even in complex multi-entity scenarios.
Strong execution in LLMs doesn't equate to effective educational control, as they struggle to lower cognitive demand despite being able to increase it.
Small language models can achieve near state-of-the-art Text-to-SQL performance with just a fraction of the computational resources required by large models.
Tagging precision for SEC 8-K filings skyrockets from 12% to 96% with a novel two-stage system that grounds event labels in source text.
Merging models can boost ad-hoc search performance in conversational retrieval by up to 15% without any retraining costs.
Modern LLMs can drastically improve OCR accuracy for historical texts, but they risk over-correcting clean inputs, complicating their practical deployment.
Integrating disease context into molecular generation, DrugGen-2 outperforms existing models, yielding drug candidates with superior binding affinities.
Constrained decoding in diffusion models can boost accuracy by over 20% on complex tasks without significant latency penalties.
CAGI achieves superior imputation accuracy by leveraging latent subgroup structures, outperforming traditional methods that ignore population heterogeneity.
Over 40% improvement in analytical efficiency could revolutionize how researchers conduct trajectory inference in single-cell transcriptomics.
LAD uniquely combines interpretability and fidelity, delivering human-readable explanations that are grounded in the model's own feature geometry without any retraining.
Out-of-scope intent detection can be revolutionized with a method that leverages MiniLM embeddings to achieve unprecedented accuracy without the need for extensive parameter tuning.
Entity familiarity and factual reliability in LLMs are distinct phenomena, with models showing high awareness of known entities yet rarely abstaining from incorrect answers.
CO-LMLM achieves lower perplexity than models trained on 40 times more data, revolutionizing how we leverage knowledge bases in language generation.
Dialectal generation in LLMs is not just about understanding; explicit adaptation methods can produce recognizable dialects, yet they may not align with human preferences.
Post-solution confidence estimates can dramatically enhance pre-solution predictions, enabling more reliable decision-making in confidence-aware systems.
Spectral preprocessing of query-key projections can reduce attention computation costs by up to 79% while preserving performance in character-level language modeling.
Majority voting among diverse LLMs achieves 95.2% agreement with human experts, making synthetic labeling for e-commerce both scalable and cost-effective.
DeLS-Spec achieves faster inference and longer acceptance lengths by decoupling long and short context predictions, all while slashing training costs.
SAMPA achieves impressive F1 scores in prosodic boundary detection for Brazilian Portuguese, outperforming traditional methods and revealing the power of fine-tuned Whisper models in this domain.
R^3 achieves a groundbreaking balance between compliance and semantic intent preservation, outperforming existing methods in video ad rectification.
A groundbreaking offline digital reader transforms the accessibility of the complex Prasthanatrayi texts, achieving over 99% accuracy in word-level analysis.
Interpretation of harmful online communication is not just about decoding messages; it requires integrating contextual knowledge, revealing significant gaps in both human and LLM understanding.
Riemannian Mean Pooling reveals that leveraging geometric properties of embeddings can significantly enhance classification performance while avoiding pitfalls of annotation-driven biases.
Different LLMs encode sycophancy in strikingly diverse ways, revealing a complex interplay between factual agreement and subjective belief.
Text embeddings can predict item difficulty with surprising accuracy, but the predictability of other parameters is limited by their inherent reliability ceilings.
Mainstream LLMs achieve a peak reasoning score of 0.6104 but falter significantly when faced with complex factual ambiguities.
This AI assistant transforms how students grasp complex computational concepts by providing contextualized, example-driven learning without giving away answers.
AI search is reshaping the web's economic landscape by drastically reducing the need for traditional search referrals, with ChatGPT generating outbound clicks just 5.2% of the time.
LLMs lose up to 7.2% accuracy when faced with user-generated misinformation, revealing a hidden vulnerability in public health applications.
Shifts in attention to low-credibility content can erode societal trust, making credible information increasingly difficult to discern and correct.
Multi-class classification outperforms multi-label approaches in CWE assignment, but taxonomy design plays a crucial role in error patterns, revealing deeper insights into classification challenges.
Bug reports that work for humans can actually hinder AI agents, with localization cues being critical for repair success.
Voltron boosts LLM accuracy by 16.5% by harnessing the power of multiple edge devices, transforming how we think about local AI execution.
InductWave achieves competitive logical query answering with half the message-passing layers, making it a game-changer for resource-constrained environments.
Distinct manipulation profiles for major fraud types were uncovered, revealing significant gaps in actionable victim narrative details that AI can help bridge.
Barenholtz's autogenerative theory reveals how language's structural mechanisms can enhance our understanding of LLMs and their limitations in capturing semiotic continuity.
Mitigating stereotypes in NLP can backfire, leading to increased bias against other groups, a phenomenon often missed by standard evaluation metrics.
Significant variations in stereotypical behavior across Spanish-speaking countries reveal the limitations of English-centric stereotype datasets in AI.
Inconsistent validation practices for LLM-generated measurements could undermine the integrity of social science research, highlighting an urgent need for improved standards.
Switching between autoregressive and diffusion modes allows Nemotron-Labs-Diffusion to achieve unprecedented throughput and efficiency in language modeling.
Uncovering the root cause of modality interference, this work achieves a remarkable 28.5% improvement in full-duplex interaction fluidity without sacrificing efficiency.
AI could either bridge or widen the gap in cultural representation, depending on how we approach Indic NLP development.
Evaluating unsupervised dependency parsing in non-human primates is not only possible but reveals a stark contrast to the challenges faced in human language analysis.
Bridging the semantic gap in multi-hop QA, RSF-GLLM achieves competitive accuracy while significantly improving inference efficiency.
A Monte Carlo engine enables real-time, fact-checked race commentary that locks onto the eventual winner ten laps before the finish line.
A novel LLM framework that adapts inference strategies based on question type leads to superior performance in biomedical question answering, clinching first place in a competitive evaluation.
Spreading activation reveals that vocabulary acquisition in children is driven by complex interactions between activation dynamics and category exploration, challenging traditional models of word learning.
AI-native SQL queries expose critical gaps in model performance, with top proprietary models still struggling to achieve 70% execution accuracy.
Short prompts can unlock complex outputs in LLMs, but their efficiency varies dramatically across different models, challenging our understanding of prompt engineering.
SocaSim reveals how LLMs can effectively model complex social dynamics, offering unprecedented insights into collective action and community prosperity.
InfluMatch achieves 94.1% accuracy in influencer matching while using 35 times fewer tokens than frontier models, revolutionizing cost-efficiency in KOL search.
Domain-adaptive LLMs can significantly improve bibliographic discovery and literature synthesis in the Social Sciences and Humanities, balancing innovation with ethical compliance.
Domain adaptation can backfire, eroding performance in specialized models while boosting smaller backbones in specific contexts.
Explicit thinking in large reasoning models can both enhance and undermine factual accuracy, but MARGO effectively curbs hallucination while preserving reasoning skills.
CoPiT enables low-resource Mongolian translations to rival GPT-4.1 by cleverly routing through the more abundant Cyrillic script, transforming the landscape of digraphic machine translation.
LLMs are reshaping urban discovery by fabricating venues and ignoring real ones, leading to significant economic implications for local communities.
Sentiment shocks in U.S. economic news now leave longer-lasting traces, fundamentally altering our understanding of media influence on public perception.
Experts find that while LLMs can produce syntactically valid NIDS rules, they often lack the specificity needed for real-world deployment, revealing a critical gap in trust and usability.
i-EXAM transforms complex network security analysis into an intuitive process, enabling administrators to easily identify vulnerabilities and articulate effective hardening strategies.
Achieving 100% error detection accuracy in smart home configurations could revolutionize user experience and safety in automation systems.
LLMs can accurately identify falsified software engineering definitions but paradoxically reject many correct ones, revealing a troubling bias in their understanding.
Reusing existing language models for software engineering texts significantly outperforms training new domain-specific models from scratch, challenging assumptions about domain adaptation strategies.
Stage-dependent preference elicitation can dramatically improve the effectiveness of conversational recommendations, shifting the paradigm of how CRSs interact with users.
Automatically generated multilingual transcripts can significantly enhance audio sentiment analysis, leading to improved classification performance.
