In this video, we're diving into some of the most noteworthy and impactful papers from this year's conference, giving you a front-row seat to the latest developments in AI.

在本視頻中，我們將深入探討今年大會上一些最值得關注、最具影響力的論文，讓您在最前排的位置瞭解人工智能的最新發展。

Let's kick things off with this paper on graph neural networks, which earned the highest review scores of the conference.

讓我們以這篇關於圖神經網絡的論文作為開場白，這篇論文獲得了本次會議最高的評審分數。

The authors identify a unifying mechanism called representation scattering that enhances various contrastive learning algorithms.

作者發現了一種名為 "表徵散射 "的統一機制，它能增強各種對比學習算法。

They propose a new framework that combines this scattering mechanism with a topology-based constraint to improve representation diversity and prevent over-scattering.

他們提出了一種新的框架，將這種散射機制與基於拓撲結構的約束相結合，以提高表示多樣性並防止過度散射。

Their benchmarks show state-of-the-art performance, solidifying this as a milestone in graph learning.

他們的基準測試結果顯示了最先進的性能，使其成為圖學習領域的一個里程碑。

Next, we have differentiable logic gate networks.

接下來是可微分邏輯門網絡。

These models use a relaxed, differentiable formulation of logic gates to achieve faster, more efficient inference compared to traditional neural networks.

與傳統的神經網絡相比，這些模型使用了一種寬鬆的、可微分的邏輯門表述，從而實現了更快、更高效的推理。

By introducing deep logic gate tree convolutions, or pooling, and residual initializations, The authors scaled these networks, achieving 86.29% accuracy on CIFAR-10 using just 61 million logic gates, being 29 times smaller than competing methods.

通過引入深度邏輯門樹卷積（或池化）和殘差初始化，作者擴大了這些網絡的規模，僅用 6100 萬個邏輯門就在 CIFAR-10 上實現了 86.29% 的準確率，是其他競爭方法的 29 倍。

We also wanted to give a shout-out to the RoadLess Scheduled, which reimagines optimization by eliminating the need for learning rate schedules, all while maintaining state-of-the-art performance across a variety of tasks.

我們還想向 RoadLess Scheduled 致敬，它通過消除對學習率計劃的需求，對優化進行了重新設計，同時在各種任務中保持了最先進的性能。

For those that seek alternatives to the transformer architecture, XLSTM introduces two variants to address the limitations of traditional LSTMs.

對於那些尋求變壓器架構替代方案的人來說，XLSTM 引入了兩種變體，以解決傳統 LSTM 的侷限性。

The SLSTM uses scalar memory and exponential gating, while the MLSTM employs matrix memory and a covariance update rule, enabling better parallelization.

SLSTM 使用標量存儲器和指數門控，而 MLSTM 則使用矩陣存儲器和協方差更新規則，從而實現了更好的並行化。

These models outperform modern alternatives like transformers and state-space models, particularly in scaling and efficiency, making them a noteworthy contender in language modeling.

這些模型優於變換器和狀態空間模型等現代模型，尤其是在擴展性和效率方面，使它們成為語言建模領域值得關注的競爭者。

Speaking of attention, Flash Attention 3 pushes the envelope with an asynchronous, low-precision mechanism that significantly speeds up attention computations on GPUs, a big step forward for efficient training and inference.

說到注意力，Flash Attention 3 採用異步、低精度機制，大大加快了 GPU 上的注意力計算速度，在高效訓練和推理方面向前邁進了一大步。

Spherical Diffusion combines a dynamics-informed diffusion framework with the Spherical Fourier Neural Operator to create highly accurate, physically consistent climate simulations.

球形擴散將動態資訊擴散框架與球形傅立葉神經運算器相結合，創建了高度精確、物理上一致的氣候模擬。

This model can emulate 100-year climate trajectories at 6 hourly intervals with minimal computational overhead, which marks a major breakthrough in climate modeling, offering stable, high-resolution simulations at a low cost.

該模型能夠以 6 小時為間隔模擬 100 年的氣候軌跡，計算開銷極低，這標誌著氣候建模領域的重大突破，能夠以較低的成本提供穩定的高分辨率模擬。

Another standout is Trajectory Flow Matching, a simulation-free approach for training neural differential equation models.

軌跡流匹配技術是另一項傑出的技術，它是一種用於訓練神經微分方程模型的免模擬方法。

This method excels at clinical time-series modeling, offering improved trajectory predictions and better uncertainty quantification.

該方法在臨床時間序列建模方面表現出色，可提供更好的軌跡預測和不確定性量化。

A team from UC Berkeley reframed humanoid control as a next-token prediction problem, similar to language modeling.

加州大學伯克利分校的一個團隊將仿人控制重構為下一個標記預測問題，類似於語言建模。

Using a causal transformer trained on diverse sensorimotor datasets, including YouTube videos, they enabled a robot to walk in real-world environments, like the streets of San Francisco, zero-shot.

他們利用在各種傳感器運動數據集（包括 YouTube 視頻）上訓練的因果轉換器，使機器人能夠在舊金山街道等真實環境中零距離行走。

On the LLM front, Row1 snagged a Best Paper award for its selective language modeling approach.

在 LLM 方面，Row1 憑藉其選擇性語言建模方法獲得最佳論文獎。

By training on the most informative tokens, rather than all tokens, it achieves state-of-the-art performance on benchmarks like math, with significantly fewer pre-training tokens.

通過對信息量最大的標記進行訓練，而不是對所有標記進行訓練，它在數學等基準測試中取得了最先進的性能，同時大大減少了預訓練標記的數量。

Special mentions go to SGLang, a system for efficiently programming complex language model workflows, and Buffer of Thoughts, a framework for reasoning that improves accuracy, efficiency, and robustness by storing high-level thought processes.

特別要提到的是 SGLang 和 Buffer of Thoughts，前者是一個用於高效編制複雜語言模型工作流的系統，後者是一個推理框架，通過存儲高級思維過程來提高準確性、效率和穩健性。

Next, DeepMind's work on many-shot in-context learning demonstrated how to leverage GemIIni's expanded context windows to incorporate hundreds or even thousands of examples.

接下來，DeepMind 在多鏡頭情境學習方面的工作展示瞭如何利用 GemIIni 的擴展情境窗口來納入數百甚至數千個示例。

Their findings showed significant performance gains across various tasks, introducing techniques like reinforced ICL and unsupervised ICL, highlighting the potential of in-context learning to rival fine-tuning in certain scenarios.

他們的研究結果表明，通過引入強化 ICL 和無監督 ICL 等技術，在各種任務中的性能都有了明顯提高，這凸顯了上下文學習在某些情況下可與微調相媲美的潛力。

Multimodality remains a hot topic, and CambrianOne steps up with a family of vision-centric multimodal large-language models.

多模態仍是一個熱門話題，CambrianOne 推出了一系列以視覺為中心的多模態大語言模型。

Using their new Spatial Vision Aggregator, the authors bridge the gap between language and vision, achieving state-of-the-art results and releasing a treasure trove of resources for the community.

作者利用他們新的空間視覺聚合器，在語言和視覺之間架起了一座橋樑，取得了最先進的成果，併為社區提供了一個資源寶庫。

On the image generation front, unlike traditional raster-scan token prediction, Visual Autoregressive Modeling uses a course-defined next-scale prediction approach, outperforming diffusion transformers on metrics like FID while being 20 times faster.

在影像生成方面，與傳統的柵格掃描標記預測不同，視覺自迴歸建模採用了航線定義的下一尺度預測方法，在 FID 等指標上優於擴散變換器，同時速度快 20 倍。

Finally, a new method for iterative reasoning optimizes chain-of-thought preferences using a refined DPO loss function with an additional negative log-likelihood term.

最後，一種用於迭代推理的新方法利用改進的 DPO 損失函數和附加的負對數可能性項優化了思維鏈偏好。

The approach significantly boosts accuracy on reasoning benchmarks like GSM 8k and math, outperforming other LLAMA2-based models.

這種方法大大提高了推理基準（如 GSM 8k 和數學）的準確性，優於其他基於 LLAMA2 的模型。

That's a wrap on our NeurIPS 2024 highlights.

以上就是 NeurIPS 2024 的精彩內容。

Did we miss a paper you think deserved the spotlight?

我們錯過了您認為值得關注的論文嗎？

Let us know in the comments below.

請在下面的評論中告訴我們。

Thanks for watching, and as always, enjoy discovery!

感謝觀看，並一如既往地享受發現的樂趣！

www.neurips.com

www.neurips.com