quill
QUILL: An Algorithm-Architecture Co-Design for Cache-Local Deformable Attention
Oh, Hyunwoo, Chen, Hanning, Yun, Sanggeon, Ni, Yang, Huang, Wenjun, Das, Tamoghno, Jang, Suyeon, Imani, Mohsen
Deformable transformers deliver state-of-the-art detection but map poorly to hardware due to irregular memory access and low arithmetic intensity. We introduce QUILL, a schedule-aware accelerator that turns deformable attention into cache-friendly, single-pass work. At its core, Distance-based Out-of-Order Querying (DOOQ) orders queries by spatial proximity; the look-ahead drives a region prefetch into an alternate buffer--forming a schedule-aware prefetch loop that overlaps memory and compute. A fused MSDeformAttn engine executes interpolation, Softmax, aggregation, and the final projection (W''m) in one pass without spilling intermediates, while small tensors are kept on-chip and surrounding dense layers run on integrated GEMMs. Implemented as RTL and evaluated end-to-end, QUILL achieves up to 7.29x higher throughput and 47.3x better energy efficiency than an RTX 4090, and exceeds prior accelerators by 3.26-9.82x in throughput and 2.01-6.07x in energy efficiency. With mixed-precision quantization, accuracy tracks FP32 within <=0.9 AP across Deformable and Sparse DETR variants. By converting sparsity into locality--and locality into utilization--QUILL delivers consistent, end-to-end speedups.
- North America > United States > California > San Francisco County > San Francisco (0.14)
- Europe > Switzerland > Zürich > Zürich (0.14)
- Asia > Japan > Honshū > Kantō > Tokyo Metropolis Prefecture > Tokyo (0.14)
- (13 more...)
Spurious Features Everywhere -- Large-Scale Detection of Harmful Spurious Features in ImageNet
Neuhaus, Yannic, Augustin, Maximilian, Boreiko, Valentyn, Hein, Matthias
Spurious Features in Training Data bird feeder graffiti eucalyptus label Benchmark performance of deep learning classifiers alone is not a reliable predictor for the performance of a deployed model. In particular, if the image classifier has picked up spurious features in the training data, its predictions can fail in unexpected ways. In this paper, we develop Hummingbird Freight Car Koala Hard Disc a framework that allows us to systematically identify Images from the web with spurious feature spurious features in large datasets like ImageNet. It is but no class features classified as class below based on our neural PCA components and their visualization. Previous work on spurious features often operates in toy settings or requires costly pixel-wise annotations. In contrast, we work with ImageNet and validate our results by showing that presence of the harmful spurious feature of a class alone is sufficient to trigger the prediction of that class. We introduce the novel dataset "Spurious ImageNet" which allows to measure the reliance of any ImageNet classifier on harmful spurious features. Moreover, we introduce SpuFix as a simple mitigation method to reduce the dependence of any ImageNet classifier on previously identified Hummingbird Freight Car Koala Hard Disc harmful spurious features without requiring additional labels Figure 1: Top: Examples of spurious features found via or retraining of the model. We provide code and data our neural PCA components but not in previous study [61].
- Leisure & Entertainment > Sports (0.93)
- Transportation > Ground > Rail (0.69)
QUILL: Query Intent with Large Language Models using Retrieval Augmentation and Multi-stage Distillation
Srinivasan, Krishna, Raman, Karthik, Samanta, Anupam, Liao, Lingrui, Bertelli, Luca, Bendersky, Mike
Large Language Models (LLMs) have shown impressive results on a variety of text understanding tasks. Search queries though pose a unique challenge, given their short-length and lack of nuance or context. Complicated feature engineering efforts do not always lead to downstream improvements as their performance benefits may be offset by increased complexity of knowledge distillation. Thus, in this paper we make the following contributions: (1) We demonstrate that Retrieval Augmentation of queries provides LLMs with valuable additional context enabling improved understanding. While Retrieval Augmentation typically increases latency of LMs (thus hurting distillation efficacy), (2) we provide a practical and effective way of distilling Retrieval Augmentation LLMs. Specifically, we use a novel two-stage distillation approach that allows us to carry over the gains of retrieval augmentation, without suffering the increased compute typically associated with it. (3) We demonstrate the benefits of the proposed approach (QUILL) on a billion-scale, real-world query understanding system resulting in huge gains. Via extensive experiments, including on public benchmarks, we believe this work offers a recipe for practical use of retrieval-augmented query understanding.
- North America > United States > New York > New York County > New York City (0.05)
- Asia > Middle East > Jordan (0.04)
Guardians of the Galaxy: where video games and Marvel truly align
It's fair to say that Square Enix didn't have the smoothest entry into the Marvel Universe. The company's Avengers-themed online action game has had problems with bugs, matchmaking and endgame repetition, and is struggling to retain an audience. But most critics agreed that its story and characterisation were strong; they just didn't belong in a live game. Guardians of the Galaxy, due out next month, is the developer's chance to redress the balance and remind Deus Ex and Tomb Raider veterans about its skill with single-player, cinematic stories. The story is classic Guardians, in that it's based around a minor misdemeanour that quickly transforms into a colossal cosmic drama.
How observability helps Quill in its mission to help kids write better
One of the concerns as schools closed at the height of the COVID-19 lockdown earlier this year was its impact on the progress of already disadvantaged pupils. Denied in-person attention as they grappled with remote learning, would they fall even further behind? The response from many teachers across the US was to turn to Quill, a non-profit organization dedicated to helping low-income students improve their writing skills, which in less than six weeks saw over a million new students sign up for its online service. With just 22 members of staff, including a six-person software engineering team, the sudden demand was a test of the organization's resilience, driving its total user population above three million. For us, that was a huge spike in new users.
- Education (1.00)
- Health & Medicine (0.77)
The Past and the Present of Natural Language Generation ai-jobs.net
Rome wasn't built in a day. It has taken years for computers to exhibit the level of intelligence they do today and be able to produce text that sounds and read human-like. It's time to appreciate this revolutionary journey. In 2007, The first step was taken by Robbie Allen, who was a veteran engineer at Cisco. He created an online college basketball website that automatically published game reviews, real-time updates, recaps, and incidents of injury.
Slack investor Index Ventures backs Slack competitor Quill – TechCrunch
Slack created a new solution for workplace communication, one copied by many, even Microsoft. But the product, which is meant to help individuals and businesses collaborate, has been critiqued for sending too many notifications, with some claiming it's sabotaged workplace productivity. Quill, a startup led by Ludwig Pettersson, Stripe's former creative director and design aficionado, claims to offer "meaningful conversations, without disturbing your team." The company has raised a $2 million seed round led by Sam Altman with participation from General Catalyst, followed by a $12.5 million Series A at a $62.5 million valuation led by Index Ventures partner and former Slack board observer Sarah Cannon, TechCrunch has learned. Quill and Cannon declined to comment.
How AI Will Make You a Better Writer.
What does it mean to read a poem, story, or novel about the human condition written entirely by a computer? This is not a crazy question. Legendary author Roald Dahl had already conjured up this nightmarish scenario for authors in one of his unnerving short stories in Someone Like You (1953). It tells the tale of the Great Automatic Grammatizator, a mammoth machine able to write prizewinning novels based on the works of living authors in 15 minutes flat. Dahl died before such a machine was within the realm of possibility.
How AI is Transforming Copywriting
Creating high-quality content, at scale, is a challenge that almost every organization faces. Larger budgets can help, but as the need for content grows, and as more channels emerge, brands are turning to artificial intelligence (AI) to help with the content creation process. AI has already made a significant impact on digital experiences, but to find out how it's changing the life of a copywriter (and a copy reader), we spoke to those in the know. Elliott Sedegah, senior product marketing manager at Adobe, claims that thanks to mounting consumer expectations, brands will be faced with increasing pressures to quickly create "personalized and relevant digital experiences." The kind of experiences that are only possible with a little help from AI. "AI is becoming your new creative assistant, and I see AI and machine learning being implemented as common tools that are specifically tasked with mundane processes that are currently holding marketers back from having sufficient time to be creative," said Sedegah.
Four companies using AI to reinvent marketing Articles Chief Digital Officer
"Its various learning mechanisms are designed around interactivity so every time a user works with Quill, it gets smarter by learning about specific domains and how to describe them," explains Katy De Leon in an interview with Marketing AI Institute. "For example, Quill can capture and integrate the nuance of how a particular company talks about its business. This type of learning results in new ways for Quill to describe and characterize the area of the business it is being asked to write about (e.g., sales performance, customer communications, employee feedback, etc.)