More Like this

1. MHC-Fine: Fine-tuned AlphaFold for precise MHC-peptide complex prediction

   https://doi.org/10.1016/j.bpj.2024.05.011  

   Glukhov, Ernest; Kalitin, Dmytro; Stepanenko, Darya; Zhu, Yimin; Nguyen, Thu; Jones, George; Patsahan, Taras; Simmerling, Carlos; Mitchell, Julie C; Vajda, Sandor; et al (September 2024, Biophysical Journal)
2. Fine-Grained Graph Rationalization

   https://doi.org/10.1145/3746252.3761307  

   Xu, Zhe; Pan, Menghai; Chen, Yuzhong; Chen, Huiyuan; Yan, Yuchen; Das, Mahashweta; Tong, Hanghang (November 2025, ACM)
3. Phases in fine volcanic ash

   https://doi.org/10.1038/s41598-023-41412-x  

   Hornby, Adrian; Gazel, Esteban; Bush, Claire; Dayton, Kyle; Mahowald, Natalie (September 2023, Scientific Reports)

   Abstract Volcanic ash emissions impact atmospheric processes, depositional ecosystems, human health, and global climate. These effects are sensitive to the size and composition of the ash; however, datasets describing the constituent phases over size ranges relevant for atmospheric transport and widely distributed impacts are practically nonexistent. Here, we present results of X-ray diffraction measurements on size-separated fractions of 40 ash samples from VEI 2–6 eruptions. We characterize changes in phase fractions with grainsize, tectonic setting, and whole-rock SiO₂. For grainsizes < 45 μm, average fractions of crystalline silica and surface salts increased while glass and iron oxides decreased with respect to the bulk sample. Samples from arc and intraplate settings are distinguished by feldspar and clinopyroxene fractions (determined by different crystallization sequences) which, together with glass, comprise 80–100% of most samples. We provide a dataset to approximate glass-free proportions of major crystalline phases; however, glass fractions are highly variable. To tackle this, we describe regressions between glass and major crystal phase fractions that help constrain the major phase proportions in volcanic ash with limited a priori information. Using our dataset, we find that pore-free ash density is well-estimated as a function of the clinopyroxene + Fe-oxide fraction, with median values of 2.67 ± 0.01 and 2.85 ± 0.03 g/cm³for intraplate and arc samples, respectively. Finally, we discuss effects including atmospheric transport and alteration on modal composition and contextualize our proximal airfall ash samples with volcanic ash cloud properties. Our study helps constrain the atmospheric and environmental budget of the phases in fine volcanic ash and their effect on ash density, integral to refine our understanding of the impact of explosive volcanism on the Earth system from single eruptions to global modeling. 

   more » « less
4. Order-Independence Without Fine Tuning

   McIlroy-Young, R; Brown, K; Olson, C; Zhang, L; Dwork, C (April 2025, ICLR)

   The development of generative language models that can create long and coherent textual outputs via autoregression has lead to a proliferation of uses and a corresponding sweep of analyses as researches work to determine the limitations of this new paradigm. Unlike humans, these ‘Large Language Models’ (LLMs) are highly sensitive to small changes in their inputs, leading to unwanted inconsistency in their behavior. One problematic inconsistency when LLMs are used to answer multiple-choice questions or analyze multiple inputs is order dependency: the output of an LLM can (and often does) change significantly when sub-sequences are swapped, despite both orderings being semantically identical. In this paper we present Set-Based Prompting, a technique that guarantees the output of an LLM will not have order dependence on a specified set of sub-sequences. We show that this method provably eliminates order dependency, and that it can be applied to any transformer-based LLM to enable text generation that is unaffected by re-orderings. Delving into the implications of our method, we show that, despite our inputs being out of distribution, the impact on expected accuracy is small, where the expectation is over the order of uniformly chosen shuffling of the candidate responses, and usually significantly less in practice. Thus, Set-Based Prompting can be used as a ‘dropped-in’ method on fully trained models. Finally, we discuss how our method’s success suggests that other strong guarantees can be obtained on LLM performance via modifying the input representations. Code is available at github.com/reidmcy/set-based-prompting. 

   more » « less
5. Antidote: Post-fine-tuning Safety Alignment for Large Language Models against Harmful Fine-tuning Attack

   Huang, Tiansheng; Bhattacharya, Gautam; Joshi, Pratik; Kimball, Josh; Liu, Ling (July 2025, IEEE)

   Safety aligned Large Language Models (LLMs) are vulnerable to harmful fine-tuning attacks (Qi et al., 2023)– a few harmful data mixed in the fine-tuning dataset can break the LLMs’s safety alignment. While several defenses have been proposed, our evaluation shows that existing defenses fail when some specific training hyper-parameters are chosen – a large learning rate or a large number of training epochs in the fine-tuning stage can easily invalidate the defense. To this end, we propose Antidote, a post-fine-tuning stage solution, which remains agnostic to the training hyper-parameters in the fine-tuning stage. Antidote relies on the philosophy that by removing the harmful parameters, the harmful model can be recovered from the harmful behaviors, regardless of how those harmful parameters are formed in the fine-tuning stage. With this philosophy, we introduce a one-shot pruning stage after harmful fine-tuning to remove the harmful weights that are responsible for the generation of harmful content. Despite its embarrassing simplicity, empirical results show that Antidote can reduce harmful score while maintaining accuracy on downstream tasks. 

   more » « less