Meet up with Cohere and learn from top NLP experts at industry and research conferences, webinars, and hackathons. 

Events - AI Conferences, NLP Webinars, LLM Hackathons | Cohere

Events

Empirical Methods in Natural Language Processing (EMNLP) is a leading conference in the area of natural language processing and artificial intelligence. 

The 

2023 Conference on Empirical Methods in Natural Language Processing

. December 6 –10. Resorts World Convention Centre

Come meet the Cohere and Cohere For AI team at Booth 13!

Check out our papers, posters, speaking engagements, and appearances by Cohere and Cohere For AI.



Maintrack Publications

Friday, December 8



Biomedical Named Entity Recognition via Dictionary-based Synonym Generalization

Specialist or Generalist? Instruction Tuning for Specific NLP Tasks

BLESS: Benchmarking Large Language Models on Sentence Simplification

, Tannon Kew, Alison Chi, Laura Vásquez-Rodríguez, Sweta Agrawal, 

, Fernando Alva-Manchego, Matthew Shardlow. 

On the Challenges of Using Black-Box APIs for Toxicity Evaluation in Research

Luiza Amador Pozzobon*, Beyza Ermis*, Patrick Lewis*, Sara Hooker*. 

NLP Evaluation in trouble: On the Need to Measure LLM Data Contamination for each Benchmark

, Iker García-Ferrero, Julen Etxaniz, Oier Lopez de Lacalle, Eneko Agirre. 

Cross-lingual Transfer Can Worsen Bias in Sentiment Analysis

Goodtriever: Adaptive Toxicity Mitigation with Retrieval-augmented Models

Locally Differentially Private Document Generation Using Zero Shot Prompting

On the State of German (Abstractive) Text Summarization

, Jing Fan and Michael Gertz. GEM Workshop Paper (extended Abstract).

11:00-12:30, Elo Uncovered: Robustness and Best Practices in Language Model Evaluation, 

Meriem Boubdir*, Edward Kim*, Beyza Ermis*, Sara Hooker* and Marzieh Fadaee*

Analyzing Multi-Sentence Aggregation in Abstractive Summarization via the Shapley Value, 

 Meng Cao, Jackie Chi Kit Cheung. NewSumm Workshop.



Delphi, and Whether Machines Can Learn Morality, 

(moderator), Liwei Jiang, Zeerak Talat. The Big Picture: Crafting a Research Narrative Workshop. 

Cohere and/or Cohere For AI team member. 



EMNLP 

The 2023 Conference on Empirical Methods in Natural Language Processing. December 6 –10. Resorts World Convention Centre

Speaker Intro: Liangyu Chen is a Research Engineer at MMLab@NTU, supervised by Prof Ziwei Liu. His research lies in data-efficient and scalable large multimodal models for human-like recognition, reasoning, and communications. He develops data collection pipelines, representations, and models for interactive or assistive AI systems. He also worked with Bloomberg Distinguished Professor Dr. Alan Yuille and Dr. Zongwei Zhou at Johns Hopkins University, on human-in-the-loop machine learning for medical imaging. Before that, he has also spent time at ByteDance, Alibaba, National University of Singapore, and A*STAR Singapore.

Cohere For AI -Guest Speaker: Liangyu Chen,Research Engineer@MMLab@NTU

Cohere for AI -  Liangyu Chen presents "Large Language Models are Visual Reasoning Coordinators"

: Lucas grew up in Belgium wanting to make video games and their AI, went on to study mechanical engineering at RWTH Aachen in Germany, did a PhD in robotic perception/computer vision there too, and is now researching representation learning and vision backbones at Google DeepMind/Brain in Zürich. Website: 

: We propose a simple pairwise Sigmoid loss for Language-Image Pre-training (SigLIP). Unlike standard contrastive learning with softmax normalization, the sigmoid loss operates solely on image-text pairs and does not require a global view of the pairwise similarities for normalization. The sigmoid loss simultaneously allows further scaling up the batch size, while also performing better at smaller batch sizes. Combined with Locked-image Tuning, with only four TPUv4 chips, we train a SigLiT model that achieves 84.5% ImageNet zero-shot accuracy in two days. The disentanglement of the batch size from the loss further allows us to study the impact of examples vs pairs and negative to positive ratio. Finally, we push the batch size to the extreme, up to one million, and find that the benefits of growing batch size quickly diminish, with a more reasonable batch size of 32k being sufficient. 

Cohere For AI - Guest Speaker: Lucas Beyer,Researcher @Google DeepMind

Cohere for AI -  Lucas Beyer presents Sigmoid Loss for Language Image Pre-Training(Geo Asia) 

Neural Information Processing Systems Foundation

 is a non-profit corporation that aims to foster the exchange of research advances in Artificial Intelligence and Machine Learning, principally by hosting an annual interdisciplinary academic conference with the highest ethical standards for a diverse and inclusive community. 
 
Join Cohere and Cohere For AI at our booth, 1109!

Check out our papers, posters, speaking engagements, and appearances by Cohere and Cohere For AI. 



, Niklas Muennighoff, Alexander M. Rush, Boaz Barak, Teven Le Scao, 

, Nouamane Tazi, Sampo Pyysalo, Thomas Wolf, Colin Raffel. 

The Goldilocks of Pragmatic Understanding: Fine-Tuning Strategy Matters for Implicature Resolution by LLMs

, Tim Rocktäschel, Edward Grefenstette. 

The Grand Illusion: The Myth of Software Portability and Implications for ML Progress

, Fraser Mince, Dzung Dinh, Jonas Kgomo, Neil Thompson and 

Location: Great Hall & Hall B1+B2 #1501



Intriguing Properties of Quantization at Scale

Arash Ahmadian*, Saurabh Dash*, Hongyu Chen*, Bharat Venkitesh*, 

, Phil Blunsom*, Ahmet Üstün*, Sara Hooker*. 

MEGABYTE: Predicting Million-byte Sequences with Multiscale Transformers

, Colin Flaherty, Armen Aghajanyan, Luke Zettlemoyer, Mike Lewis.

 Location: Great Hall & Hall B1+B2 #541



Repetition In Repetition Out: Towards Understanding Neural Text Degeneration from the Data Perspective

, Huayang Li, Tian Lan, Zihao Fu, Deng Cai, Lemao Liu, Nigel Collier, Taro Watanabe, 

Improving Language Plasticity via Pretraining with Active Forgetting

, Roberta Raileanu, David Ifeoluwa Adelani, Pontus Stenetorp, Sebastian Riedel, Mikel Artetxe. 

D4: Improving LLM Pretraining via Document De-Duplication and Diversification

FAIR-Ensemble: Homogeneous Deep Ensembling Naturally Attenuates Disparate Group Performances

, Daniel Dsouza, Karina Nguyen, Randall Balestriero, 

. Workshop on Algorithmic Fairness through the Lens of Time. 

When Less is More: Investigating Data Pruning for Pretraining LLMs at Scale

, Ahmet Üstün*, Luiza A Pozzobon*, Alex Wang*, Marzieh Fadaee*, Sara Hooker*. 

Workshop on Attributing Model Behaviour at Scale (ATTRIB). 

Generalizable Agents for Neural Network Optimisation

, Kale-ab Tessera, Callum Tilbury, Sasha Abramowitz, Ruan de Kock W, Omayma Mahjoub, Benjamin Rosman,

, Arnu Pretorius. Workshop on Advancing Neural Network Training (WANT): Computational Efficiency, Scalability, and Resource Optimization. 



10:30-11:00, Beyond the Research Paper, 

9:30-10:00, Avoiding AI's "Moore's Law": Why we are building a ladder to the moon. The Future of Attribution in ML. 

11:00-11:45, Avoiding AI's "Moore's Law": Why we are building a ladder to the moon. The Future of Attribution in ML. Panelists: 



14:00-14:30, AYA: Multilingual Instruction Tuning and Open Challenges, 

. Instruction Tuning and Instruction Following.

16:15-17:15,

Funding and institutional support for research at the intersection of machine learning and physical sciences. Panelists: 

 Jesse Thaler, Max Welling. ML and the Physical Sciences Workshop. 

10:30-11:30, Panel on "Socially Responsible Language Modelling Research”. Moderator: 

, Panelists: Roger Grosse, Vinodkumar Prabhakaran, Stella Biderman, David Bau. SoLaR workshop



*Cohere and/or Cohere For AI team member

NeurIPS

Cohere and Cohere For AI will be at NeurIPS from December 10-16

 Yang Chen is a Ph.D. student at Georgia Tech, supervised by Professors Alan Ritter and Wei Xu. His research interests lie in grounding the visual knowledge of multimodal large language models through retrieval-augmented generation. He has interned at Google DeepMind, working with Hexiang Hu and Ming-Wei Chang on visual entity reasoning using multimodal LLMs. Prior to that, he received an M.S. in Computer Science from the University of Chicago and a Bachelor's degree from the University of Melbourne.

Multimodal Large Language Models (MLLMs) have demonstrated state-of-the-art capabilities in various tasks involving both images and text, including visual question answering. However, it remains unclear whether these MLLMs possess the ability to answer information-seeking queries of an image such as 'When was this church built?'.

In this talk, I will first introduce InfoSeek, a dataset tailored for visual information-seeking questions that cannot be answered using only common sense knowledge. I will then present insights into the generalization and instruction-tuning of MLLMs using InfoSeek. Finally, I will discuss what the future holds for multimodal retrieval models and how MLLMs-powered generative search engines could transform the existing search experiences.

https://open-vision-language.github.io/infoseek/

Cohere For AI - Guest Speaker: Yang Chen, Ph.D. student @ Georgia Tech

Cohere for AI -  Yang Chen presents "Visual Knowledge Grounding for Multimodal Large Language Models"

Join us on Dec 15 to close this epic chapter of Aya, Cohere For AI's open-science initiative to unlock multilingual generative AI! 

To mark the last day to contribute to our open-source multilingual dataset, we're hosting a 4-hour live stream featuring lightning talks & panels from collaborators worldwide. 

Full schedule and links to join, coming soon!

Join the movement 

Aya Grande Finale

A 4-hour live stream to celebrate all the work that has gone into Aya, Cohere For AI's open-science initiative to unlock multilingual generative AI.

Join the Geo Regional Asia group as they welcome Haihao Shen to present "Efficient LLM Inference on CPUs."

Bio: Haihao is a senior AI architect in DCAI/AISE at Intel, leading model quantization and efficient inference for LLMs on Intel platforms.

Description: the talk will give an overview of LLM model quantization and efficient inference based on Intel Extension for Transformers.

Cohere For AI - Guest Speaker: Haihao Shen, Sr. AI architect @ Intel

Cohere for AI -  Haihao Shen presents Efficient LLM Inference on CPUs (Geo-Asia)

Bio: Jie Huang is a Ph.D. candidate at UIUC working on various aspects of large language models, including factuality, reasoning, and safety. His research aims to develop powerful and responsible large language models and apply them to solve real-world problems. (

)

Abstract: Large Language Models (LLMs) have emerged as a groundbreaking technology with their unparalleled text generation capabilities across various applications. Nevertheless, concerns persist regarding the accuracy and appropriateness of their generated content. A contemporary methodology, self-correction, has been proposed as a remedy to these issues. Building upon this premise, we critically examine the role and efficacy of self-correction within LLMs, shedding light on its true potential and limitations. Central to our investigation is the notion of intrinsic self-correction, whereby an LLM attempts to correct its initial responses based solely on its inherent capabilities, without the crutch of external feedback. In the context of reasoning, our research indicates that LLMs struggle to self-correct their responses without external feedback, and at times, their performance might even degrade post self-correction. Drawing from these insights, we offer suggestions for future research and practical applications in this field. (

Cohere For AI - Guest Speaker: Jie Huang, Ph.D. candidate at UIUC

Cohere for AI -  Jie Huang presents Large Language Models Cannot Self-Correct Reasoning Yet

Description:
Transformers have recently emerged as a powerful tool for learning visual representations. In this paper, we identify and characterize artifacts in feature maps of both supervised and self-supervised ViT networks. The artifacts correspond to high-norm tokens appearing during inference primarily in low-informative background areas of images, that are repurposed for internal computations. We propose a simple yet effective solution based on providing additional tokens to the input sequence of the Vision Transformer to fill that role. We show that this solution fixes that problem entirely for both supervised and self-supervised models, sets a new state of the art for self-supervised visual models on dense visual prediction tasks, enables object discovery methods with larger models, and most importantly leads to smoother feature maps and attention maps for downstream visual processing.
abs: https://arxiv.org/abs/2309.16588v1

Intro: I am a PhD student at FAIR Paris (Meta) and INRIA Grenoble, co-advised by Piotr Bojanowski, Maxime Oquab, Julien Mairal and (formerly) Armand Joulin. My work focuses on self-supervised learning (DINOv2) and understanding ViTs (vision transformers need registers). Prior to that, I studied at École polytechnique, and at ENS (MVA master).

Cohere For AI - Guest Speaker: Timothée Darcet, PhD @ Meta AI & INRIA

Cohere for AI -  Timothée Darcet presents Vision Transformers need Registers (Geo Asia)

 "Arthur is a Senior Research Scientist at DeepMind, he got his PhD in 2022 from Sorbonne, Paris, for his work on Continual Learning. Now living in London, he's working on making IA training more collaborative through distributed training. @Ar_Douillard on Twitter." 


 Large language models (LLM) have become a critical component in many applications of machine learning. However, standard approaches to training LLM require a large number of tightly interconnected accelerators, with devices exchanging gradients and other intermediate states at each optimization step. While it is difficult to build and maintain a single computing cluster hosting many accelerators, it might be easier to find several computing clusters each hosting a smaller number of devices. In this work, we propose a distributed optimization algorithm, Distributed Low-Communication (DiLoCo), that enables training of language models on islands of devices that are poorly connected. The approach is a variant of federated averaging, where the number of inner steps is large, the inner optimizer is AdamW, and the outer optimizer is Nesterov momentum. On the widely used C4 dataset, we show that DiLoCo on 8 workers performs as well as fully synchronous optimization while communicating 500 times less. DiLoCo exhibits great robustness to the data distribution of each worker. It is also robust to resources becoming unavailable over time, and vice versa, it can seamlessly leverage resources that become available during training. abs: 

Cohere For AI - Guest Speaker:Arthur Douillard, Sr. Research @DeepMind

Cohere for AI -  Arthur Douillard presents DiLoCo: Distributed Low-Communication Training of Language Models (Geo Asia)

 I am a PhD student in Computer Science at the University of Washington advised by Prof. Luke Zettlemoyer and Prof. Noah A. Smith. I have been a visiting research at Meta AI, working with Scott Yih. Prior to UW, I graduated from UCLA with a B.S. in Computer Science and Minor in Math. Twitter: 

 Large language models (LMs) are currently trained to predict tokens given document prefixes, enabling them to directly perform long-form generation and prompting-style tasks which can be reduced to document completion. Existing pretraining pipelines train LMs by concatenating random sets of short documents to create input contexts but the prior documents provide no signal for predicting the next document. We instead present In-Context Pretraining, a new approach where language models are pretrained on a sequence of related documents, thereby explicitly encouraging them to read and reason across document boundaries. We can do In-Context Pretraining by simply changing the document ordering so that each context contains related documents, and directly applying existing pretraining pipelines. However, this document sorting problem is challenging. There are billions of documents and we would like the sort to maximize contextual similarity for every document without repeating any data. To do this, we introduce approximate algorithms for finding related documents with efficient nearest neighbor search and constructing coherent input contexts with a graph traversal algorithm.abs: 

Cohere For AI - Guest Speaker: Weijia Shi, PhD student at UW

Cohere for AI -  Weijia Shi presents In-Context Pretraining: Language Modeling Beyond Document Boundaries

EMNLP

Upcoming Events

Cohere’s past events