Search papers, labs, and topics across Lattice.
100 papers published across 3 labs.
Super-spreader traits are the key drivers behind the most influential misinformation spreaders, revealing critical insights for targeted interventions.
Current benchmarks may mislead researchers about the necessity of complex modeling, as simple recency-weighted methods outperform advanced architectures in several cases.
CoRRe achieves superior recommendation performance by seamlessly integrating collaborative signals into LLM-generated item representations, all without any training.
SSR-GRPO significantly reduces noise in retrieval systems, enhancing relevance assessments and enabling more accurate e-commerce search results.
Users show markedly different levels of acceptance for their own AI agents compared to those of others, revealing critical insights into the dynamics of AI integration in online dating.
Current benchmarks may mislead researchers about the necessity of complex modeling, as simple recency-weighted methods outperform advanced architectures in several cases.
CoRRe achieves superior recommendation performance by seamlessly integrating collaborative signals into LLM-generated item representations, all without any training.
SSR-GRPO significantly reduces noise in retrieval systems, enhancing relevance assessments and enabling more accurate e-commerce search results.
Users show markedly different levels of acceptance for their own AI agents compared to those of others, revealing critical insights into the dynamics of AI integration in online dating.
Falcon-7B with Summarize Chains not only automates financial news summarization but does so with remarkable accuracy, outperforming traditional methods and exposing critical weaknesses in RAG techniques.
LFU emerges as the clear champion among eviction policies, with alternatives failing to deliver meaningful improvements in cache performance.
$TCP_\alpha$ guarantees complete separation of confidence scores for correct and incorrect predictions, transforming how we trust model outputs in music information retrieval.
PETA achieves superior ligand ranking by adapting pretrained models at test time with only 0.03% of the parameters updated, revolutionizing efficiency in virtual screening.
Achieving state-of-the-art zero-shot recommendation performance, RecPFN redefines efficiency in sequential recommendation systems by leveraging synthetic data and causal priors.
DraftFM achieves over 50% accuracy in predicting draft picks before any cards are actually drafted, revolutionizing day-zero drafting strategies.
Efficiently routing queries in AI systems can be achieved without the costly overhead of exhaustive value estimation, thanks to novel policies that balance accuracy and cost.
Generative modeling can revolutionize ride-hailing dispatch by directly optimizing driver-passenger assignments, leading to unprecedented service quality improvements.
Fine-tuning on synthetic data allows for a dramatic boost in code retrieval performance, achieving a balanced macro nDCG@10 of 0.5992 in a challenging domain.
Blockchain technology can effectively filter out malicious nodes in mobile edge caching while incentivizing resource sharing among users.
Intermediate context compression can cut GPU energy usage by over 50% while maintaining quality, challenging static compression strategies in edge RAG applications.
TrustRAG transforms RAG systems by enabling verifiable trust in document retrieval through a decentralized, expert-driven scoring process.
SCoRD transforms LLM-based recommendation systems by enabling retrievers to adapt intelligently to user intent without the heavy cost of frequent LLM updates.
Shifting the evaluation of agentic search to a vast, uncurated corpus reveals a dramatic decline in retrieval effectiveness, challenging current models' capabilities.
Madhhab-aware filtering can more than double retrieval accuracy for school-specific fiqh questions, revealing a crucial dimension in Islamic jurisprudence retrieval.
IAR achieves a remarkable boost in retrieval-free question answering, outperforming traditional methods by effectively internalizing document knowledge into LLMs.
SCORE achieves a remarkable 53.23% Top-1 accuracy in EEG-to-image retrieval, outperforming existing methods by over 17 percentage points, even without target labels.
Anchoring neural and visual representations can boost brain-to-image retrieval accuracy by over 9% with minimal stimulus repetitions, challenging the reliance on trial averaging.
Pairwise ranking not only outperforms single-action RL in explanation selection but also slashes serving costs and latency, making it a game-changer for industrial applications.
Adaptive similarity margins in HN-CLIP boost retrieval accuracy by up to 4.3% while training 2.4x faster than leading methods.
Early backpropagation can boost training throughput by over 9% without sacrificing model performance.
Pair-Aware Discriminative Reasoning in UMER reveals critical distinctions between semantically similar candidates, elevating retrieval accuracy in multimodal tasks.
AI Product Research Agents can significantly boost customer engagement and sales by delivering personalized recommendations directly through WhatsApp, transforming how e-commerce platforms interact with users.
Incomplete-knowledge robustness isn't a one-size-fits-all issue; MissDiag reveals that the type of missing evidence dramatically influences system performance.
Complete evidence retrieval can be dramatically improved by strategically allocating context, achieving a 23.8% boost without increasing the retrieval budget.
Resynthesizing audio from coarse tokens can be dramatically improved by leveraging the geometric structure of the RVQ layer, leading to better fidelity than traditional methods.
HARP outperforms traditional CVE prioritization methods by dynamically adapting to implicit operational preferences, revealing the power of leveraging historical data without explicit prompts.
Over 45% of child-targeted YouTube videos fail to disclose paid promotions, raising serious concerns about transparency and compliance.
DeepWeaver transforms the way LLMs synthesize evidence, leading to answers that are not only more comprehensive but also better grounded in citations.
A single lightweight model can now achieve superior recommendation performance by leveraging a structured memory of reasoning, eliminating the need for costly repeated computations.
LLM-based query reformulation not only bridges the gap in retrieval performance but also sets a new standard for legal question answering in Greek statutory contexts.
BFTR guarantees zero overcharging while maximizing customer utility, challenging traditional telecom pricing models.
Swapping to CTIFoundry allows smaller models to outperform flagship models, achieving higher accuracy with fewer tool calls in cyber threat intelligence investigations.
Gaussian noise isn't enough; a new method reveals that even noise-protected embeddings can leak sensitive information.
Long-horizon planning with online adaptation can significantly enhance service robots' efficiency in environments with unpredictable, time-varying rewards.
TDIR reveals that you can effectively disentangle temporal and categorical information in image embeddings, enabling unprecedented flexibility in historical image retrieval.
Interface health in generative recommendation systems is multi-signal, revealing that prefix alignment is critical for candidate exposure but weakens under certain scoring conditions.
Bridging the intent gap in e-commerce, TTP boosts order volume by 0.46% through reasoning-driven personalized retrieval.
GateDiffInt reveals how structured intent extraction can significantly boost conversion rates by effectively managing noise in user behavior data.
PILOT's proactive framework boosts recommendation system performance by over 40% in search efficiency and achieves up to 1.60% gains in core metrics, all without human oversight.
OneModel boosts user engagement by over 1% while simplifying the engineering complexity of multi-scenario ranking systems.
Technicians can now access critical maintenance information in under 12 seconds, transforming how they interact with complex manuals.
Retrieval architecture can dramatically influence AI performance, with structural failures outnumbering reasoning errors by a staggering 95 to 15 in financial reconciliation tasks.
Automating Assurance Case generation could drastically reduce the resource burden on SMEs striving for compliance with the EU Cyber Resilience Act.
Users are three times more likely to deploy their own agents than to engage with others', revealing a critical asymmetry that could hinder the effectiveness of agentic recommender systems in online dating.
COVID-19 reshaped vaccine discourse on YouTube, but the real story lies in how audience comments reveal persistent conspiracies and political tensions that outlast the pandemic.
OGR achieves a remarkable 48.2% boost in recommendation effectiveness by seamlessly integrating semantic and collaborative signals into slate generation.
Misleading attacks can exploit security copilots even with factually correct documents, revealing a critical vulnerability in RAG systems.
CABLE boosts memory retrieval performance by focusing on complementary associations, achieving higher scores in scenarios with distributed evidence across sessions.
Reader-specific preferences can enhance performance, but they don't ensure consistent intervention success across contexts—highlighting a critical gap in our understanding of evidence utility in ML systems.
Explicit domain adaptation can transform molecular language models from inconsistent performers to top-tier representations in targeted discovery tasks.
Keyword searches on TikTok reveal up to 56% harmful content, significantly outpacing passive scrolling results and challenging existing moderation assumptions.
CoAL-RAG achieves a remarkable 42.5% boost in BLEU scores for legal question retrieval, redefining efficiency and interpretability in complex legal consultations.
Digital item theft drives a significant portion of cybercrime in gaming, revealing how game mechanics can create unintended vulnerabilities for players.
GBP not only enhances ranking accuracy in noisy environments but also gracefully handles increased complexity, outperforming traditional methods in real-world scenarios.
OVIP-SG outperforms existing frameworks by preserving small object instances and enhancing retrieval accuracy, revolutionizing the mapping of fine-grained objects in 3D environments.
Multimodal embeddings not only streamline asset personalization at Netflix but also deliver a dramatic boost in performance for cold-start scenarios and video previews.
Agents can significantly enhance retrieval performance in visually rich document environments, achieving a 67.50% Recall@1 compared to just 37.50% for OCR-text methods.
DEPT achieves superior retrieval quality by preserving document embeddings while adapting query expansions, revealing a new synergy between these traditionally separate tasks.
FLEXRec shows that compact LLMs can achieve state-of-the-art recommendation accuracy without the computational burden of larger models.
Removing AI features from search results boosts publisher traffic, revealing a troubling trade-off between AI integration and user trust.
International crises can synchronize political narratives across media, but national policies still drive significant divergence in coverage.
MemTree3D slashes the computational cost of 3D question answering, boosting performance while avoiding the inefficiencies of traditional visual search methods.
OnGameLearn not only navigates the complexities of strategic interactions but also adapts to evolving contextual signals, achieving superior performance in competitive pricing scenarios.
Polaris shows that optimizing LLM-generated table descriptions for retrieval effectiveness can lead to significant performance improvements over existing methods.
Reasoning-capable models can significantly reduce the impact of misinformation in RAG systems without the heavy computational costs of isolation.
Wallets with large token holdings can still be flagged as risky if their stablecoin reserves don't cover loan sizes, revealing critical liquidity mismatches in decentralized lending.
Sellers can now optimize pricing strategies with a new FPTAS that achieves revenue maximization despite incomplete knowledge of buyer preferences.
Sampling experts from a multinomial policy can significantly enhance routing efficiency while satisfying operational constraints, achieving low regret in real-world applications.
GRIP cuts query-latent mutual information by 30x while slashing hallucinations by 73%, revolutionizing how we leverage retrieved evidence in reasoning tasks.
Hyper-M2RAG redefines multimodal retrieval by capturing complex relationships in a hypergraph structure, achieving superior performance with less computational overhead.
Incremental updates to semantic substrates can be 33.7 times cheaper than full re-computation, challenging the assumption that corpus size dictates maintenance costs.
Index-side pruning can slash retrieval latency by up to 6.6x across diverse engines, while query pruning is largely redundant in modern systems.
Dynamic scaling in D2-ScaleAgent allows for adaptive retrieval and reasoning, leading to superior long document comprehension compared to traditional fixed workflows.
The best commercially-licensed retrieval system still trails behind its free counterpart, revealing a hidden "commercial tax" that could impact enterprise decisions.
No existing speech retrieval method can robustly handle the full spectrum of user intents, revealing a critical gap in current technologies.
LENS achieves 84.8% evidence recall without the overhead of pre-materializing evidence, redefining how we approach dynamic document retrieval.
Simpler chunking methods often outperform complex ones, challenging the assumption that more sophisticated approaches yield better retrieval results.
Grafting trainable components onto frozen teacher models reveals the hidden pitfalls of generalization in sequential recommenders, leading to substantial performance gains.
Evidence lineages in LineageRAG not only improve retrieval accuracy but also ensure that every piece of evidence is grounded in verifiable source text.
DSPrompt reshapes retrieval semantics to thwart adversarial attacks without altering the retrieval pipeline, achieving robust defense with less than 1% additional parameters.
IGD can boost factual accuracy in language models by over 65% while ensuring they still follow user context effectively.
Super-spreader traits are the key drivers behind the most influential misinformation spreaders, revealing critical insights for targeted interventions.
GEO-optimized content is more prevalent than expected, with nearly 9% of web pages showing signs of manipulation, raising alarms about the integrity of information in generative search engines.
Zero-shot cyber threat detection can be dramatically improved by leveraging structured risk indicators generated from user activity timelines, outperforming previous state-of-the-art methods.
Ingestion-time defenses against coordinated poisoning are fundamentally flawed, allowing attackers to manipulate retrieval systems with minimal effort.
Predicting impression distribution before online evaluation can drastically improve the effectiveness of ranking models, revealing hidden shifts that traditional metrics overlook.
UniDot's architecture unifies feature interactions and sequential modeling, achieving state-of-the-art performance in industrial recommendation tasks with unprecedented efficiency.
A novel dataset that distinguishes between complementary and related products could revolutionize how e-commerce platforms enhance user experience and product discovery.
Privacy-preserving RFANNS can now be performed on encrypted vector databases without compromising the efficiency of search queries.
Position and popularity biases in recommender systems can be effectively mitigated, leading to recommendations that truly reflect user preferences rather than skewed interactions.
Neglecting the structural differences in candidate negatives can lead to suboptimal recommendations, but SAHC-NS adapts to these variations, enhancing sample quality and model performance.
Source-style collapse can cause fine-tuned retrievers to miss relevant capabilities, but a simple TF-IDF signal can dramatically improve retrieval success rates.
Domain-specific triplet fine-tuning can dramatically enhance the performance of embedding models in distinguishing between similar entity records, achieving substantial gains in retrieval accuracy.
When index recall falls below 0.5, exposure and churn can skyrocket, rendering traditional admission checks ineffective against retrieval hubs.
LLM-MGCL achieves a staggering 52% boost in recommendation recall by effectively integrating LLM-derived semantic insights with traditional collaborative signals.
