AGF-TI: Dual Roles in ML & Condensed Matter
- The paper proposes a graph-based semi-supervised learning method that uses adversarial fusion, tensorial imputation, and anchor alignment to recover incomplete multi-view data.
- AGF-TI in condensed matter describes a dual-topological state in monolayer AgF2 where a single ferroelastic/polar distortion induces both electronic and magnonic non-trivial topology.
- Both applications leverage efficient optimization and structured design to address data fragmentation in machine learning and to achieve symmetry-driven topological effects in physics.
AGF-TI has two distinct meanings in recent research. In machine learning, it denotes “Adversarial Graph Fusion for Incomplete Multi-view Semi-supervised Learning with Tensorial Imputation,” a method for incomplete multi-view semi-supervised learning that combines adversarial graph fusion, tensorial recovery of missing graph structure, and anchor-based label propagation (Jiang et al., 19 Sep 2025). In condensed-matter physics, the same label denotes the dual-topological character of monolayer AgF, where a single ferroelastic/polar distortion simultaneously generates non-trivial electronic and magnonic topology in a relativistic altermagnet (González et al., 21 Aug 2025).
1. Terminological scope
The acronym is therefore domain-dependent rather than universally standardized. The two principal usages represented in the literature considered here are summarized below.
| Usage | Domain | Defining content |
|---|---|---|
| AGF-TI | Machine learning | Adversarial graph fusion with tensorial imputation for incomplete multi-view semi-supervised learning |
| AGF-TI | Condensed matter | Dual topology in monolayer AgF: fermionic high-Chern subbands plus a Chern magnon insulator |
In the machine-learning usage, AGF-TI is a graph-based semi-supervised framework designed for view-missing data, with explicit notation for views, samples, anchors, labeled samples, and classes. In the AgF usage, “AGF-TI” denotes the dual topological character of monolayer AgF (“AgF”), with “TI” used in a generalized “topological index/insulator” sense rather than as a time-reversal-invariant topological insulator (Jiang et al., 19 Sep 2025, González et al., 21 Aug 2025).
This dual usage is not merely lexical. The machine-learning construction addresses robustness under incomplete observations and distorted local geometry, whereas the AgF0 construction concerns a symmetry-driven coexistence of bosonic and fermionic topological invariants in a single polar altermagnetic phase. The shared acronym therefore masks two unrelated but internally coherent technical programs.
2. AGF-TI in incomplete multi-view semi-supervised learning
In the machine-learning literature, AGF-TI is formulated for incomplete multi-view semi-supervised learning, where each view has features 1, the first 2 samples are labeled, and the one-hot label matrix satisfies 3. The method is built around per-view sample-anchor bipartite graphs 4, anchor-alignment matrices 5 with 6, and a consensus fused bipartite graph 7 constrained to the row-simplex domain 8 (Jiang et al., 19 Sep 2025).
Its central motivation is the “Sub-Cluster Problem” (SCP). SCP refers to the fragmentation of what should be a single manifold cluster into disconnected sub-clusters because missing samples create “vacuum regions” in a view. The stated consequence is that the classical smoothness assumption for label propagation is violated: labels cannot diffuse smoothly across the vacuum region, and graph fusion becomes distorted because per-view graphs omit missing nodes and misrepresent the underlying manifold geometry. This is the specific failure mode AGF-TI is designed to alleviate (Jiang et al., 19 Sep 2025).
The fused sample-anchor graph is defined as
9
with normalized Laplacian
0
Within this graph, semi-supervised learning is expressed through a Laplacian smoothness objective. In the single-view form given in the source,
1
where 2 contains soft labels and 3 is a diagonal fitting-weight matrix. This smoothness-based formulation is then lifted to the fused sample-anchor graph used by AGF-TI (Jiang et al., 19 Sep 2025).
3. Adversarial graph fusion, tensorial imputation, and anchors
The adversarial graph-fusion component learns a robust consensus bipartite graph from incomplete per-view graphs through a min-max formulation. The operator is
4
where 5 are per-view weights on the simplex and 6 regularizes the connectivity of 7 (Jiang et al., 19 Sep 2025).
The full AGF-TI objective couples adversarial fusion, label propagation, and tensorial imputation: 8 with 9. Here 0 balances adversarial fusion and label propagation, while 1 controls tensorial imputation strength (Jiang et al., 19 Sep 2025).
Tensorial imputation is the mechanism used to recover missing sample-anchor relations. The per-view bipartite graphs are stacked into a third-order tensor
2
whose low-rank structure is enforced through the tensor nuclear norm based on t-SVD. The stated rationale is that “high-order consistency” across anchors, views, and samples enables recovery of incomplete graph structure that is invisible to viewwise methods. This makes imputation a joint component of the optimization rather than a preprocessing step (Jiang et al., 19 Sep 2025).
Anchors provide the computational reduction. Instead of building dense sample-sample graphs, AGF-TI operates directly on sample-anchor bipartite graphs. The anchor construction problem is
3
For missing samples, the rows 4 are initialized to equal probabilities 5 and later updated within the alternating scheme. This anchor-based strategy reduces the stated time complexity from 6 to 7 and the space complexity from 8 to 9, with 0 (Jiang et al., 19 Sep 2025).
4. Optimization, empirical behavior, and limitations
The optimization is an alternating procedure with an inner min-max problem in 1 and outer ADMM-style updates for 2, 3, 4, and the auxiliary tensor variable 5. The inner objective is
6
The paper states that 7 is convex in 8, that the gradient is
9
and that reduced gradient descent with Armijo line search converges to the global optimum of the inner minimization over 0 (Jiang et al., 19 Sep 2025).
For label propagation, the block variable 1 is updated by
2
with a Woodbury-based block inversion that lowers the principal cost of the update to 3. The paper further states that, under a mild “sufficient inner optimization” assumption, the outer ADMM iterations yield bounded sequences and the overall algorithm converges to a stationary KKT point of the full objective (Jiang et al., 19 Sep 2025).
Empirically, AGF-TI is evaluated on CUB, UCI-Digit, Caltech101-20, OutScene, MNIST-USPS, and AwA, using View Missing Ratio (VMR) in 4, Label Annotation Ratio (LAR) in 5, and the metrics ACC, Precision, and F1. Each experiment is repeated 10 times. Under LAR6, the reported highlights include UCI-Digit ACC values of 7, 8, and 9 at VMR 0, 1, and 2, respectively; CUB ACC values of 3, 4, and 5; OutScene ACC values of 6, 7, and 8; and MNIST-USPS ACC values of 9, 0, and 1 (Jiang et al., 19 Sep 2025).
The ablation results identify three components as consequential: anchor alignment 2, view weights 3, and tensorial imputation. On Caltech101-20 and OutScene, removing 4 causes large drops; a cited example is ACC decreasing from 5 to 6 on Caltech101-20 at VMR 7. Tensorial imputation becomes especially important at high missing rates, where removing TI causes ACC to drop by up to approximately 8 at VMR 9 versus approximately 0 at VMR 1. The reported robust parameter ranges are 2 and 3, with 4 fixed to 5 in the experiments. Runtime examples at VMR6, LAR7 are 8 s for CUB, 9 s for UCI-Digit, 0 s for Caltech101-20, 1 s for OutScene, 2 s for MNIST-USPS, and 3 s for AwA (Jiang et al., 19 Sep 2025).
The stated limitations are equally specific. Anchors are pre-selected rather than jointly optimized with fusion and imputation; the convergence proof assumes sufficiently accurate inner optimization; gains from tensorial imputation may diminish when cross-view correlations are weak; and the tensor nuclear norm step can be heavy for extremely large 4 or 5. These caveats place AGF-TI within the broader class of structured but computationally nontrivial graph-based SSL methods (Jiang et al., 19 Sep 2025).
5. AGF-TI as dual topology in monolayer AgF6
In condensed-matter usage, AGF-TI refers to monolayer AgF7 as a “dual-topological” altermagnet, meaning that both electronic and magnonic excitations acquire non-trivial Chern topology through the same ferroelastic/polar distortion (González et al., 21 Aug 2025). The high-symmetry reference structure is a flat centrosymmetric monolayer with space group 8 (No. 123), whereas the ground state is a tilted polar structure with space group 9 (No. 4). The order parameter 00 is quantified by out-of-plane tilt 01, in-plane rotation 02, and vertical buckling 03 Å, with in-plane anisotropy 04. The flat reference is metastable and higher in energy by approximately 05 eV/f.u. (González et al., 21 Aug 2025).
The magnetic state is a collinear, compensated altermagnet in the nonrelativistic limit, with zero net moment, while spin-orbit coupling induces a tiny weak-ferromagnetic canting of approximately 06 per formula unit. The distortion breaks inversion and lowers rotational symmetry while retaining a nonsymmorphic 07 screw along the crystallographic 08 axis. Without SOC, the system exhibits even-parity, momentum-symmetric spin splitting 09, with anisotropy of approximately 10 meV along 11–12–13 and vanishing splitting along 14–15. With SOC, a multipolar analysis gives an overall balance 16 and 17, with 18-wave strongest in the 19 and 20 channels and 21-wave dominant along 22 (González et al., 21 Aug 2025).
This structural and magnetic setting is essential because the same distortion activates both topological sectors. The paper describes the mechanism as follows: inversion breaking and SOC open avoided electronic crossings and enable strong Dzyaloshinskii–Moriya interaction on the distorted bonds, so the single structural order parameter 23 simultaneously topologizes electrons and magnons. In this usage, AGF-TI is therefore a material classification rather than an algorithmic framework (González et al., 21 Aug 2025).
6. Electronic and magnonic topology, control, and broader AgF24 context
The electronic sector of monolayer AgF25 is not classified as a time-reversal-invariant 26 topological insulator, because time-reversal symmetry is broken by magnetic order, and it is not yet an intrinsic quantum anomalous Hall insulator in its pristine ground state because the net occupied Chern number cancels. Instead, the top two valence bands carry opposite high Chern numbers,
27
so the total occupied valence-manifold Chern number remains 28. SOC opens avoided crossings up to approximately 29 meV near 30 and 31, where the Berry curvature concentrates. The paper states that if a perturbation such as uniaxial strain, an out-of-plane electric field, a staggered substrate potential, or appropriate heterostructure coupling isolates one of these bands and places 32 in the Chern gap, a QAH state with
33
should be obtainable, with three co-propagating electronic edge channels per boundary (González et al., 21 Aug 2025).
The magnon sector is already topological in the distorted ground state. In the flat 34 reference, DMI is symmetry-forbidden and the magnon branches are degenerate and topologically trivial. In the polar 35 phase with SOC, representative exchange parameters are 36 meV, 37 meV, 38 meV, with 39–40 meV on dominant bonds and symmetric anisotropy up to approximately 41. Linear spin-wave theory and bosonic BdG analysis yield two non-degenerate magnon branches with a full topological gap and Chern numbers
42
Bulk-boundary correspondence then implies one chiral magnon edge mode per boundary traversing the gap, together with a finite transverse thermal Hall conductivity 43 below the estimated Néel temperature 44–45 K (González et al., 21 Aug 2025).
The unifying control parameter is ferroelastic/polar domain switching. Reversing the polar axis, 46, reverses the relevant symmetry-allowed SOC and DMI vectors, flips the signs of both the electronic and magnonic Chern indices, and reverses the anomalous charge and heat Hall responses. The paper identifies the transverse thermal Hall effect as the clearest macroscopic fingerprint of the magnon Chern bands, while ribbon spectroscopy or multi-terminal transport would be natural probes of a strain- or field-induced 47 QAH state. Chiral magnon edge modes are proposed to be accessible by Brillouin light scattering or nonlocal magnon transport, and RIXS is identified as a probe of both multi-magnon dynamics and electronic band splittings (González et al., 21 Aug 2025).
Broader AgF48 chemistry provides context for why this topological proposal is notable. Independent spectroscopy and cluster/DFT analysis identify AgF49 as a charge-transfer insulator closely analogous to undoped La50CuO51, with optical interband absorption onset at approximately 52 eV, a pronounced charge-transfer peak centered at about 53 eV, an estimated optical fundamental gap 54 eV, and F K-edge RIXS 55 excitations at 56, 57, and 58 eV (Bachar et al., 2021). Related theoretical work on Ag59OF60 explicitly describes AgF61 as an antiferromagnetic charge-transfer positive-62 insulator with Ag(II) 63 character (Domański et al., 2021). This suggests that the dual-topological monolayer proposal is embedded in a broader silver-fluoride materials landscape already known for strong covalency, charge-transfer physics, and unusual competing phases.